diff --git a/.github/workflows/run-tests.yml b/.github/workflows/run-tests.yml index 8a0f812..9651a54 100644 --- a/.github/workflows/run-tests.yml +++ b/.github/workflows/run-tests.yml @@ -46,13 +46,12 @@ jobs: - name: Install dependencies run: | python -m pip install --upgrade pip - pip install tox coverage + pip install "tox<4.24" coverage - name: Run tests run: >- - pipx run --python '${{ steps.setup-python.outputs.python-path }}' tox - -- -rFEx --durations 10 --color yes --cov --cov-branch --cov-report=xml # pytest args + -- -rFEx --durations 10 --color yes --cov --cov-branch --cov-report=xml - name: Check for codecov token availability id: codecov-check diff --git a/docs/tutorials/2-variant-effect-prediction.ipynb b/docs/tutorials/2-variant-effect-prediction.ipynb index fb61bce..3777367 100644 --- a/docs/tutorials/2-variant-effect-prediction.ipynb +++ b/docs/tutorials/2-variant-effect-prediction.ipynb @@ -16,7 +16,7 @@ }, { "cell_type": "code", - "execution_count": 1, + "execution_count": 2, "metadata": {}, "outputs": [], "source": [ @@ -43,7 +43,7 @@ }, { "cell_type": "code", - "execution_count": 2, + "execution_count": 3, "metadata": {}, "outputs": [ { @@ -74,30 +74,63 @@ " genes and variants mapped based on these column not based on the genomic\n", " locus based on the annotaiton.\n", "\n", + " >>> decima vep -v \"data/sample.vcf\" -o \"vep_results.parquet\" --distance-\n", + " type tss --min-distance 50000 --max-distance 100000 # predict for\n", + " variants within 50kb of the TSS and 100kb of the TSS\n", + "\n", + " >>> decima vep -v \"data/sample.vcf\" -o \"vep_results.parquet\" --save-\n", + " replicates # save the replicates in the output parquet file\n", + "\n", + " >>> decima vep -v \"data/sample.vcf\" -o \"vep_results.parquet\" --genome\n", + " \"hg38\" # use hg38 genome build\n", + "\n", + " >>> decima vep -v \"data/sample.vcf\" -o \"vep_results.parquet\" --genome\n", + " \"path/to/fasta/hg38.fa\" # use custom genome build\n", + "\n", "Options:\n", - " -v, --variants PATH Path to the variant file .vcf file\n", - " -o, --output_pq PATH Path to the output parquet file.\n", - " --tasks TEXT Tasks to predict. If not provided, all tasks will\n", - " be predicted.\n", - " --chunksize INTEGER Number of variants to process in each chunk.\n", - " Loading variants in chunks is more memory\n", - " efficient.This chuck of variants will be process\n", - " and saved to output parquet file before contineus\n", - " to next chunk. Default: 10_000.\n", - " --model INTEGER Model to use for variant effect prediction either\n", - " replicate number or path to the model.\n", - " --device TEXT Device to use. Default: None which automatically\n", - " selects the best device.\n", - " --batch-size INTEGER Batch size for the model. Default: 1.\n", - " --num-workers INTEGER Number of workers for the loader. Default: 1.\n", - " --max-distance FLOAT Maximum distance from the TSS. Default: 524288.\n", - " --max-distance-type TEXT Type of maximum distance. Default: tss.\n", - " --include-cols TEXT Columns to include in the output in the original\n", - " tsv file to include in the output parquet file.\n", - " Default: None.\n", - " --gene-col TEXT Column name for gene names. Default: None.\n", - " --genome TEXT Genome build. Default: hg38.\n", - " --help Show this message and exit.\n", + " -v, --variants PATH Path to the variant .vcf file. VCF file needs to\n", + " be normalized. Try normalizing th vcf file in\n", + " case of an error. `bcftools norm -f ref.fasta\n", + " input.vcf.gz -o output.vcf.gz`\n", + " -o, --output_pq PATH Path to the output parquet file.\n", + " --tasks TEXT Tasks to predict. If not provided, all tasks will\n", + " be predicted.\n", + " --chunksize INTEGER Number of variants to process in each chunk.\n", + " Loading variants in chunks is more memory\n", + " efficient.This chuck of variants will be process\n", + " and saved to output parquet file before contineus\n", + " to next chunk. Default: 10_000.\n", + " --model TEXT `0`, `1`, `2`, `3`, `ensemble` or a path or a\n", + " comma-separated list of paths to safetensor files\n", + " to perform variant effect prediction. Default:\n", + " `ensemble`.\n", + " --metadata TEXT Path to the metadata anndata file or name of the\n", + " model. If not provided, the compabilite metadata\n", + " for the model will be used. Default: ensemble.\n", + " --device TEXT Device to use. Default: None which automatically\n", + " selects the best device.\n", + " --batch-size INTEGER Batch size for the model. Default: 8\n", + " --num-workers INTEGER Number of workers for the loader. Default: 4\n", + " --distance-type TEXT Type of distance. Default: tss.\n", + " --min-distance FLOAT Minimum distance from the end of the gene.\n", + " Default: 0.\n", + " --max-distance FLOAT Maximum distance from the TSS. Default: 524288.\n", + " --include-cols TEXT Columns to include in the output in the original\n", + " tsv file to include in the output parquet file.\n", + " Default: None.\n", + " --gene-col TEXT Column name for gene names. Default: None.\n", + " --genome TEXT Genome build. Default: hg38.\n", + " --save-replicates Save the replicates in the output parquet file.\n", + " Default: False. Only supported for ensemble\n", + " models.\n", + " --disable-reference-cache Disables the reference cache which significantly\n", + " speeds up the computation by caching the\n", + " reference expression predictios in the metadata.\n", + " --float-precision TEXT Floating-point precision to be used in\n", + " calculations. Avaliable options include:\n", + " '16-true', '16-mixed', 'bf16-true', 'bf16-mixed',\n", + " '32-true', '64-true', '32', '16', and 'bf16'.\n", + " --help Show this message and exit.\n", "\u001b[0m" ] } @@ -131,16 +164,36 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 4, "metadata": {}, - "outputs": [], + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "decima - INFO - Using device: 0 and genome: hg38\n", + "Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.\n", + "decima - INFO - Performing predictions on VariantDataset(48 variants from ['chr1'] between 516455 and 110234512 bp from TSS)\n", + "/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/lightning_fabric/plugins/environments/slurm.py:204: PossibleUserWarning: The `srun` command is available on your system but is not used. HINT: If your intention is to run Lightning on SLURM, prepend your python command with `srun` like so: srun python /gpfs/scratchfs01/site/u/lala8/conda/envs/decima/bin ...\n", + "GPU available: True (cuda), used: True\n", + "TPU available: False, using: 0 TPU cores\n", + "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n", + "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n", + "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n", + "\u001b[2KPredicting \u001b[35m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m 96/96 \u001b[2m0:01:56 • 0:00:00\u001b[0m \u001b[2;4m0.71it/s\u001b[0m it/s\u001b[0m it/s\u001b[0m \n", + "\u001b[?25hdecima - WARNING - Warnings:\n", + "decima - WARNING - allele_mismatch_with_reference_genome: 10 alleles out of 48 predictions mismatched with the genome file /home/lala8/.local/share/genomes/hg38/hg38.fa. If this is not expected, please check if you are using the correct genome version.\n", + "\u001b[0m" + ] + } + ], "source": [ "! decima vep -v \"data/sample.vcf\" -o \"vep_vcf_results.parquet\"" ] }, { "cell_type": "code", - "execution_count": 4, + "execution_count": 5, "metadata": {}, "outputs": [ { @@ -148,17 +201,17 @@ "output_type": "stream", "text": [ "unknown: 0 / 48 \n", - "allele_mismatch_with_reference_genome: 26 / 48 \n" + "allele_mismatch_with_reference_genome: 10 / 48 \n" ] } ], "source": [ - "! cat vep_vcf_results.parquet.warnings.log" + "! cat vep_vcf_results.warnings.log" ] }, { "cell_type": "code", - "execution_count": 5, + "execution_count": 6, "metadata": {}, "outputs": [ { @@ -219,16 +272,16 @@ " 163840\n", " 172672\n", " ...\n", - " -0.000053\n", - " -0.000153\n", - " -0.000089\n", - " -0.000016\n", - " -0.000013\n", - " -0.000040\n", - " -0.000070\n", - " 7.629395e-05\n", - " -0.000026\n", - " -0.000069\n", + " -0.038816\n", + " -0.132628\n", + " 0.012945\n", + " -0.057034\n", + " -0.045095\n", + " -0.019845\n", + " -0.048861\n", + " -0.248287\n", + " -0.157361\n", + " -0.202191\n", " \n", " \n", " 1\n", @@ -243,16 +296,16 @@ " 163840\n", " 178946\n", " ...\n", - " -0.000586\n", - " -0.000891\n", - " -0.000565\n", - " -0.000311\n", - " -0.000461\n", - " -0.000252\n", - " -0.000376\n", - " -6.060898e-04\n", - " -0.000454\n", - " -0.000725\n", + " 0.007422\n", + " -0.040294\n", + " 0.037184\n", + " -0.039102\n", + " -0.069273\n", + " -0.089758\n", + " 0.085668\n", + " -0.100051\n", + " 0.199948\n", + " -0.228135\n", " \n", " \n", " 2\n", @@ -267,16 +320,16 @@ " 163840\n", " 170729\n", " ...\n", - " -0.000565\n", - " -0.000787\n", - " -0.000515\n", - " -0.000279\n", - " -0.000354\n", - " -0.000202\n", - " -0.000342\n", - " -5.399883e-04\n", - " -0.000423\n", - " -0.000597\n", + " -0.038910\n", + " -0.132768\n", + " 0.012589\n", + " -0.057297\n", + " -0.044415\n", + " -0.019266\n", + " -0.049203\n", + " -0.247513\n", + " -0.158220\n", + " -0.202740\n", " \n", " \n", " 3\n", @@ -291,16 +344,16 @@ " 163840\n", " 165050\n", " ...\n", - " -0.001453\n", - " -0.001775\n", - " -0.001403\n", - " -0.000820\n", - " -0.000896\n", - " -0.000575\n", - " -0.001113\n", - " -1.119316e-03\n", - " -0.001036\n", - " -0.001202\n", + " 0.006907\n", + " -0.041188\n", + " 0.036539\n", + " -0.038984\n", + " -0.070001\n", + " -0.090065\n", + " 0.085498\n", + " -0.100311\n", + " 0.199196\n", + " -0.229159\n", " \n", " \n", " 4\n", @@ -315,16 +368,16 @@ " 163840\n", " 166306\n", " ...\n", - " 0.000045\n", - " 0.000105\n", - " 0.000060\n", - " -0.000030\n", - " -0.000049\n", - " 0.000029\n", - " 0.000135\n", - " 1.490116e-07\n", - " 0.000057\n", - " 0.000001\n", + " 0.384163\n", + " 0.383833\n", + " 0.208923\n", + " 0.197689\n", + " 0.147001\n", + " 0.214184\n", + " 0.429241\n", + " 0.046892\n", + " 0.232819\n", + " 0.122424\n", " \n", " \n", " 5\n", @@ -339,16 +392,16 @@ " 163840\n", " 167179\n", " ...\n", - " -0.000561\n", - " -0.000826\n", - " -0.000536\n", - " -0.000299\n", - " -0.000324\n", - " -0.000196\n", - " -0.000374\n", - " -3.589988e-04\n", - " -0.000342\n", - " -0.000426\n", + " 0.128392\n", + " 0.201155\n", + " 0.096084\n", + " 0.117298\n", + " -0.126285\n", + " -0.059694\n", + " 0.184677\n", + " -0.009753\n", + " 0.305165\n", + " -0.107976\n", " \n", " \n", " 6\n", @@ -363,16 +416,16 @@ " 163840\n", " 198383\n", " ...\n", - " -0.000399\n", - " -0.000585\n", - " -0.000374\n", - " -0.000208\n", - " -0.000212\n", - " -0.000126\n", - " -0.000265\n", - " -2.926290e-04\n", - " -0.000266\n", - " -0.000376\n", + " 0.382839\n", + " 0.387868\n", + " 0.207169\n", + " 0.192711\n", + " 0.148570\n", + " 0.212126\n", + " 0.429055\n", + " 0.040387\n", + " 0.231833\n", + " 0.117729\n", " \n", " \n", " 7\n", @@ -387,16 +440,16 @@ " 163840\n", " 184493\n", " ...\n", - " 0.001317\n", - " 0.001001\n", - " 0.000884\n", - " 0.001013\n", - " 0.000691\n", - " 0.001310\n", - " 0.001097\n", - " 5.598068e-04\n", - " 0.000501\n", - " 0.000827\n", + " 0.127002\n", + " 0.199668\n", + " 0.092870\n", + " 0.113858\n", + " -0.132513\n", + " -0.059923\n", + " 0.180629\n", + " -0.008512\n", + " 0.302857\n", + " -0.112850\n", " \n", " \n", " 8\n", @@ -411,16 +464,16 @@ " 163840\n", " 168975\n", " ...\n", - " -0.000244\n", - " -0.000214\n", - " -0.000310\n", - " -0.000153\n", - " -0.000324\n", - " -0.000168\n", - " -0.000147\n", - " -4.823208e-04\n", - " -0.000211\n", - " -0.000401\n", + " -0.065035\n", + " -0.009808\n", + " -0.089304\n", + " -0.123990\n", + " -0.237687\n", + " -0.145043\n", + " -0.027265\n", + " -0.172405\n", + " -0.209000\n", + " -0.169588\n", " \n", " \n", " 9\n", @@ -435,16 +488,16 @@ " 163840\n", " 173223\n", " ...\n", - " 0.000012\n", - " -0.000113\n", - " 0.000002\n", - " -0.000028\n", - " -0.000248\n", - " -0.000283\n", - " -0.000314\n", - " -1.890063e-04\n", - " -0.000022\n", - " -0.000029\n", + " 0.149715\n", + " 0.036652\n", + " 0.236477\n", + " -0.033082\n", + " -0.059123\n", + " -0.109544\n", + " -0.116550\n", + " -0.056748\n", + " 0.238428\n", + " -0.066090\n", " \n", " \n", " 10\n", @@ -459,16 +512,16 @@ " 163840\n", " 170339\n", " ...\n", - " -0.000286\n", - " -0.000486\n", - " -0.000312\n", - " -0.000150\n", - " -0.000256\n", - " -0.000124\n", - " -0.000179\n", - " -3.057718e-04\n", - " -0.000183\n", - " -0.000252\n", + " -0.064135\n", + " -0.008839\n", + " -0.088697\n", + " -0.123552\n", + " -0.236969\n", + " -0.144438\n", + " -0.026115\n", + " -0.171792\n", + " -0.208494\n", + " -0.170012\n", " \n", " \n", " 11\n", @@ -483,16 +536,16 @@ " 163840\n", " 177242\n", " ...\n", - " 0.007558\n", - " 0.004934\n", - " 0.007966\n", - " 0.007595\n", - " -0.001943\n", - " 0.001598\n", - " 0.007838\n", - " 2.528191e-03\n", - " 0.007137\n", - " 0.009534\n", + " 0.149233\n", + " 0.036425\n", + " 0.236163\n", + " -0.033192\n", + " -0.059114\n", + " -0.109789\n", + " -0.116972\n", + " -0.057177\n", + " 0.238166\n", + " -0.067322\n", " \n", " \n", " 12\n", @@ -507,16 +560,16 @@ " 163840\n", " 166924\n", " ...\n", - " -0.000190\n", - " -0.000248\n", - " -0.000215\n", - " -0.000113\n", - " -0.000070\n", - " -0.000076\n", - " -0.000159\n", - " -5.897880e-05\n", - " -0.000117\n", - " -0.000182\n", + " 0.025078\n", + " 0.316090\n", + " 0.052550\n", + " -0.128108\n", + " -0.025624\n", + " 0.018586\n", + " 0.070953\n", + " 0.035494\n", + " 0.035036\n", + " 0.198598\n", " \n", " \n", " 13\n", @@ -531,16 +584,16 @@ " 163840\n", " 166653\n", " ...\n", - " -0.000217\n", - " -0.000332\n", - " -0.000225\n", - " -0.000132\n", - " -0.000127\n", - " -0.000102\n", - " -0.000146\n", - " -2.127588e-04\n", - " -0.000168\n", - " -0.000276\n", + " -0.060781\n", + " 0.136466\n", + " 0.222455\n", + " 0.180580\n", + " -0.077465\n", + " 0.059501\n", + " -0.152240\n", + " 0.021208\n", + " 0.388104\n", + " 0.088033\n", " \n", " \n", " 14\n", @@ -555,16 +608,16 @@ " 163840\n", " 199838\n", " ...\n", - " -0.000643\n", - " -0.000886\n", - " -0.000513\n", - " -0.001688\n", - " -0.000346\n", - " -0.000528\n", - " -0.000894\n", - " -4.293919e-04\n", - " -0.001309\n", - " -0.000301\n", + " 0.024326\n", + " 0.315336\n", + " 0.050626\n", + " -0.121845\n", + " -0.024810\n", + " 0.020802\n", + " 0.065568\n", + " 0.039766\n", + " 0.038300\n", + " 0.207067\n", " \n", " \n", " 15\n", @@ -579,16 +632,16 @@ " 163840\n", " 178999\n", " ...\n", - " -0.000031\n", - " -0.000034\n", - " -0.000048\n", - " -0.000017\n", - " 0.000019\n", - " -0.000002\n", - " -0.000036\n", - " -1.719594e-05\n", - " -0.000026\n", - " -0.000037\n", + " -0.069126\n", + " 0.131456\n", + " 0.215256\n", + " 0.164175\n", + " -0.088568\n", + " 0.049765\n", + " -0.160795\n", + " 0.012626\n", + " 0.377364\n", + " 0.080307\n", " \n", " \n", " 16\n", @@ -603,16 +656,16 @@ " 163840\n", " 168116\n", " ...\n", - " -0.000396\n", - " -0.000607\n", - " -0.000421\n", - " -0.000306\n", - " -0.000294\n", - " -0.000207\n", - " -0.000237\n", - " -3.867447e-04\n", - " -0.000289\n", - " -0.000498\n", + " -0.068721\n", + " -0.131200\n", + " -0.095311\n", + " 0.046062\n", + " 0.058227\n", + " -0.014569\n", + " -0.110806\n", + " 0.156508\n", + " 0.050760\n", + " 0.086102\n", " \n", " \n", " 17\n", @@ -627,16 +680,16 @@ " 163840\n", " 183816\n", " ...\n", - " -0.000809\n", - " -0.001129\n", - " -0.000884\n", - " -0.000664\n", - " -0.000741\n", - " -0.000528\n", - " -0.000512\n", - " -7.908940e-04\n", - " -0.000607\n", - " -0.000990\n", + " -0.000722\n", + " -0.065477\n", + " -0.015448\n", + " -0.048512\n", + " 0.080478\n", + " 0.044108\n", + " -0.040190\n", + " -0.109651\n", + " 0.016822\n", + " -0.062620\n", " \n", " \n", " 18\n", @@ -651,16 +704,16 @@ " 163840\n", " 181059\n", " ...\n", - " -0.000826\n", - " -0.001292\n", - " -0.000932\n", - " -0.000710\n", - " -0.000789\n", - " -0.000506\n", - " -0.000457\n", - " -9.316206e-04\n", - " -0.000666\n", - " -0.001122\n", + " -0.068234\n", + " -0.130586\n", + " -0.095205\n", + " 0.046179\n", + " 0.059056\n", + " -0.013978\n", + " -0.110189\n", + " 0.157281\n", + " 0.050832\n", + " 0.085974\n", " \n", " \n", " 19\n", @@ -675,16 +728,16 @@ " 163840\n", " 176946\n", " ...\n", - " -0.000033\n", - " -0.000067\n", - " -0.000013\n", - " -0.000010\n", - " 0.000030\n", - " 0.000017\n", - " -0.000020\n", - " 1.147389e-04\n", - " -0.000005\n", - " -0.000017\n", + " -0.000299\n", + " -0.064919\n", + " -0.015113\n", + " -0.048239\n", + " 0.080818\n", + " 0.044669\n", + " -0.040059\n", + " -0.109231\n", + " 0.017035\n", + " -0.062525\n", " \n", " \n", " 20\n", @@ -699,16 +752,16 @@ " 163840\n", " 177982\n", " ...\n", - " -0.000297\n", - " -0.000456\n", - " -0.000343\n", - " -0.000253\n", - " -0.000310\n", - " -0.000204\n", - " -0.000172\n", - " -3.470182e-04\n", - " -0.000220\n", - " -0.000371\n", + " 0.144722\n", + " 0.096918\n", + " 0.053467\n", + " 0.178367\n", + " 0.186895\n", + " -0.086145\n", + " 0.078297\n", + " 0.339905\n", + " 0.164013\n", + " 0.351803\n", " \n", " \n", " 21\n", @@ -723,16 +776,16 @@ " 163840\n", " 172928\n", " ...\n", - " -0.000118\n", - " -0.000153\n", - " -0.000130\n", - " -0.000090\n", - " -0.000128\n", - " -0.000076\n", - " -0.000053\n", - " -2.347827e-04\n", - " -0.000090\n", - " -0.000214\n", + " -0.204408\n", + " -0.164630\n", + " -0.161219\n", + " -0.077524\n", + " 0.001738\n", + " -0.079267\n", + " -0.283000\n", + " -0.061010\n", + " -0.083843\n", + " -0.061747\n", " \n", " \n", " 22\n", @@ -747,16 +800,16 @@ " 163840\n", " 180515\n", " ...\n", - " 0.000099\n", - " 0.000077\n", - " 0.000057\n", - " 0.000040\n", - " 0.000018\n", - " 0.000047\n", - " 0.000075\n", - " 2.750456e-04\n", - " 0.000065\n", - " 0.000258\n", + " 0.145138\n", + " 0.099315\n", + " 0.055320\n", + " 0.177228\n", + " 0.190383\n", + " -0.084195\n", + " 0.083160\n", + " 0.344412\n", + " 0.166400\n", + " 0.354686\n", " \n", " \n", " 23\n", @@ -771,16 +824,16 @@ " 163840\n", " 180515\n", " ...\n", - " 0.003284\n", - " 0.005105\n", - " 0.002662\n", - " 0.001423\n", - " 0.001525\n", - " 0.001257\n", - " 0.001919\n", - " 5.420089e-03\n", - " 0.002166\n", - " 0.005263\n", + " -0.216383\n", + " -0.174637\n", + " -0.171691\n", + " -0.086247\n", + " -0.017797\n", + " -0.098000\n", + " -0.294861\n", + " -0.081399\n", + " -0.096354\n", + " -0.078279\n", " \n", " \n", " 24\n", @@ -795,16 +848,16 @@ " 163840\n", " 179428\n", " ...\n", - " 0.002200\n", - " 0.001744\n", - " 0.001206\n", - " 0.001263\n", - " 0.001008\n", - " -0.000117\n", - " 0.001218\n", - " 2.799034e-04\n", - " 0.001503\n", - " 0.001372\n", + " 0.125082\n", + " 0.187445\n", + " 0.094098\n", + " 0.073541\n", + " 0.153524\n", + " 0.122184\n", + " 0.178186\n", + " 0.325020\n", + " 0.131711\n", + " 0.251710\n", " \n", " \n", " 25\n", @@ -819,16 +872,16 @@ " 163840\n", " 179428\n", " ...\n", - " -0.004052\n", - " -0.003465\n", - " -0.003092\n", - " -0.003744\n", - " -0.003384\n", - " -0.003265\n", - " -0.003253\n", - " 1.600981e-03\n", - " -0.001915\n", - " -0.002728\n", + " -0.072443\n", + " -0.081550\n", + " -0.038486\n", + " -0.047714\n", + " 0.226176\n", + " 0.045261\n", + " -0.126760\n", + " 0.371777\n", + " 0.060598\n", + " -0.005164\n", " \n", " \n", " 26\n", @@ -843,16 +896,16 @@ " 163840\n", " 173165\n", " ...\n", - " 0.000001\n", - " 0.000010\n", - " 0.000091\n", - " -0.000079\n", - " -0.000044\n", - " -0.000061\n", - " -0.000042\n", - " -4.082918e-05\n", - " -0.000022\n", - " -0.000060\n", + " 0.125406\n", + " 0.187519\n", + " 0.094387\n", + " 0.074029\n", + " 0.154567\n", + " 0.122893\n", + " 0.178522\n", + " 0.326261\n", + " 0.131973\n", + " 0.251846\n", " \n", " \n", " 27\n", @@ -867,16 +920,16 @@ " 163840\n", " 173165\n", " ...\n", - " 0.000536\n", - " -0.000179\n", - " 0.001585\n", - " 0.001431\n", - " 0.003103\n", - " 0.001631\n", - " 0.000257\n", - " 2.081454e-03\n", - " 0.000913\n", - " 0.001677\n", + " -0.072282\n", + " -0.081517\n", + " -0.038386\n", + " -0.047584\n", + " 0.226377\n", + " 0.045594\n", + " -0.126571\n", + " 0.372053\n", + " 0.060832\n", + " -0.004993\n", " \n", " \n", " 28\n", @@ -891,16 +944,16 @@ " 163840\n", " 172720\n", " ...\n", - " 0.000584\n", - " 0.000577\n", - " 0.000674\n", - " 0.000536\n", - " 0.000167\n", - " 0.000397\n", - " 0.000507\n", - " 1.705885e-04\n", - " 0.000503\n", - " 0.000481\n", + " -0.109378\n", + " -0.095939\n", + " -0.046820\n", + " -0.160233\n", + " -0.108172\n", + " -0.116325\n", + " 0.049603\n", + " 0.171065\n", + " 0.111874\n", + " 0.250584\n", " \n", " \n", " 29\n", @@ -915,16 +968,16 @@ " 163840\n", " 172720\n", " ...\n", - " 0.003762\n", - " 0.002849\n", - " 0.001487\n", - " 0.002733\n", - " -0.000099\n", - " 0.000631\n", - " 0.001469\n", - " 4.820824e-04\n", - " 0.002098\n", - " -0.000244\n", + " -0.014295\n", + " -0.047262\n", + " 0.040566\n", + " 0.034287\n", + " 0.086552\n", + " -0.002746\n", + " -0.036511\n", + " 0.105897\n", + " 0.075976\n", + " -0.094442\n", " \n", " \n", " 30\n", @@ -939,16 +992,16 @@ " 163840\n", " 172374\n", " ...\n", - " 0.000155\n", - " 0.000251\n", - " 0.000114\n", - " 0.000134\n", - " -0.000137\n", - " -0.000068\n", - " 0.000125\n", - " 4.252791e-05\n", - " 0.000117\n", - " 0.000137\n", + " -0.114771\n", + " -0.101728\n", + " -0.051283\n", + " -0.167664\n", + " -0.115162\n", + " -0.118221\n", + " 0.045587\n", + " 0.164287\n", + " 0.105892\n", + " 0.242462\n", " \n", " \n", " 31\n", @@ -963,16 +1016,16 @@ " 163840\n", " 172374\n", " ...\n", - " 0.000258\n", - " 0.000810\n", - " 0.000308\n", - " 0.000489\n", - " 0.000618\n", - " 0.000236\n", - " 0.000523\n", - " 6.991625e-04\n", - " 0.000155\n", - " 0.000536\n", + " -0.014773\n", + " -0.051328\n", + " 0.039865\n", + " 0.034642\n", + " 0.079301\n", + " -0.005801\n", + " -0.037709\n", + " 0.102587\n", + " 0.075023\n", + " -0.093098\n", " \n", " \n", " 32\n", @@ -987,16 +1040,16 @@ " 163840\n", " 170946\n", " ...\n", - " -0.000651\n", - " -0.000626\n", - " -0.000446\n", - " -0.000290\n", - " -0.000174\n", - " -0.000026\n", - " -0.000154\n", - " -3.945976e-04\n", - " -0.000396\n", - " -0.000499\n", + " 0.017597\n", + " -0.080504\n", + " 0.049995\n", + " -0.011808\n", + " -0.271627\n", + " -0.061634\n", + " -0.068092\n", + " -0.169892\n", + " -0.139209\n", + " -0.134266\n", " \n", " \n", " 33\n", @@ -1011,16 +1064,16 @@ " 163840\n", " 170946\n", " ...\n", - " 0.014822\n", - " 0.029157\n", - " 0.015316\n", - " 0.007980\n", - " 0.015211\n", - " 0.007929\n", - " 0.010251\n", - " 2.181430e-02\n", - " 0.009214\n", - " 0.024373\n", + " -0.297318\n", + " 0.036640\n", + " 0.039409\n", + " 0.172793\n", + " 0.131960\n", + " 0.099102\n", + " -0.127269\n", + " 0.248293\n", + " 0.078142\n", + " 0.244484\n", " \n", " \n", " 34\n", @@ -1035,16 +1088,16 @@ " 163840\n", " 233549\n", " ...\n", - " 0.002052\n", - " 0.002637\n", - " 0.003581\n", - " 0.001084\n", - " -0.001867\n", - " -0.003804\n", - " 0.003472\n", - " -1.005411e-03\n", - " 0.003400\n", - " 0.001397\n", + " 0.017213\n", + " -0.081016\n", + " 0.048742\n", + " -0.012799\n", + " -0.271882\n", + " -0.061404\n", + " -0.068229\n", + " -0.168541\n", + " -0.140811\n", + " -0.135465\n", " \n", " \n", " 35\n", @@ -1059,16 +1112,16 @@ " 163840\n", " 233549\n", " ...\n", - " -0.002514\n", - " -0.002176\n", - " 0.001273\n", - " 0.007376\n", - " 0.011271\n", - " 0.019305\n", - " -0.002886\n", - " 2.035785e-02\n", - " -0.001660\n", - " 0.006990\n", + " -0.296310\n", + " 0.027849\n", + " 0.034702\n", + " 0.172035\n", + " 0.133321\n", + " 0.101695\n", + " -0.133212\n", + " 0.247434\n", + " 0.074210\n", + " 0.243297\n", " \n", " \n", " 36\n", @@ -1083,16 +1136,16 @@ " 163840\n", " 179789\n", " ...\n", - " 0.000870\n", - " 0.001771\n", - " 0.002003\n", - " 0.000641\n", - " -0.002239\n", - " -0.000491\n", - " 0.001610\n", - " -6.132126e-04\n", - " 0.002603\n", - " 0.000714\n", + " 0.173036\n", + " 0.238832\n", + " -0.112437\n", + " -0.396939\n", + " -0.048346\n", + " -0.349238\n", + " 0.178761\n", + " -0.541497\n", + " -0.432859\n", + " -0.115770\n", " \n", " \n", " 37\n", @@ -1107,16 +1160,16 @@ " 163840\n", " 179789\n", " ...\n", - " 0.015270\n", - " 0.004848\n", - " 0.011452\n", - " 0.028524\n", - " 0.009958\n", - " 0.013799\n", - " 0.007597\n", - " 2.641273e-02\n", - " 0.009691\n", - " 0.026180\n", + " -0.587394\n", + " -0.383097\n", + " -0.074972\n", + " 0.116038\n", + " -0.106850\n", + " -0.536714\n", + " -0.289871\n", + " 0.275144\n", + " -0.386951\n", + " -0.475129\n", " \n", " \n", " 38\n", @@ -1131,16 +1184,16 @@ " 163840\n", " 182577\n", " ...\n", - " -0.000336\n", - " 0.001399\n", - " 0.000511\n", - " 0.000450\n", - " -0.000501\n", - " -0.000243\n", - " -0.001328\n", - " 4.479885e-04\n", - " 0.000544\n", - " 0.002848\n", + " 0.196702\n", + " 0.244239\n", + " -0.096241\n", + " -0.383522\n", + " -0.053860\n", + " -0.346986\n", + " 0.204916\n", + " -0.547306\n", + " -0.428087\n", + " -0.109396\n", " \n", " \n", " 39\n", @@ -1155,16 +1208,16 @@ " 163840\n", " 182577\n", " ...\n", - " -0.032543\n", - " -0.033684\n", - " -0.029430\n", - " -0.050647\n", - " -0.030112\n", - " -0.021851\n", - " -0.012187\n", - " -2.968431e-02\n", - " -0.037312\n", - " -0.067244\n", + " -0.598326\n", + " -0.350383\n", + " -0.062013\n", + " 0.137351\n", + " -0.113740\n", + " -0.549226\n", + " -0.316263\n", + " 0.282046\n", + " -0.367199\n", + " -0.457674\n", " \n", " \n", " 40\n", @@ -1179,16 +1232,16 @@ " 163840\n", " 205369\n", " ...\n", - " -0.001145\n", - " -0.001371\n", - " 0.000107\n", - " -0.000453\n", - " -0.001597\n", - " -0.000798\n", - " -0.001882\n", - " 1.063347e-04\n", - " 0.000106\n", - " 0.000086\n", + " -0.065015\n", + " -0.134536\n", + " -0.140023\n", + " 0.024319\n", + " -0.060507\n", + " -0.108153\n", + " -0.149791\n", + " -0.038775\n", + " -0.076944\n", + " -0.118178\n", " \n", " \n", " 41\n", @@ -1203,16 +1256,16 @@ " 163840\n", " 205369\n", " ...\n", - " 0.006410\n", - " 0.010421\n", - " 0.008430\n", - " 0.010849\n", - " -0.004994\n", - " 0.003090\n", - " -0.004843\n", - " -9.270430e-03\n", - " 0.008755\n", - " 0.011827\n", + " -0.138112\n", + " -0.135524\n", + " -0.122887\n", + " 0.003460\n", + " -0.095921\n", + " -0.142851\n", + " -0.133408\n", + " -0.044298\n", + " -0.079864\n", + " 0.063350\n", " \n", " \n", " 42\n", @@ -1227,16 +1280,16 @@ " 163840\n", " 185065\n", " ...\n", - " 0.000849\n", - " 0.001993\n", - " 0.001033\n", - " -0.000050\n", - " -0.000534\n", - " 0.000095\n", - " 0.000409\n", - " -3.244877e-04\n", - " 0.000842\n", - " 0.000388\n", + " -0.064936\n", + " -0.134381\n", + " -0.140410\n", + " 0.024382\n", + " -0.060211\n", + " -0.106950\n", + " -0.150407\n", + " -0.038308\n", + " -0.077518\n", + " -0.118339\n", " \n", " \n", " 43\n", @@ -1251,16 +1304,16 @@ " 163840\n", " 185065\n", " ...\n", - " 0.009973\n", - " 0.012482\n", - " 0.003073\n", - " 0.019083\n", - " 0.010606\n", - " 0.000387\n", - " 0.003512\n", - " 7.757962e-03\n", - " 0.011860\n", - " 0.019928\n", + " -0.138833\n", + " -0.136745\n", + " -0.123534\n", + " 0.002528\n", + " -0.096342\n", + " -0.143056\n", + " -0.133687\n", + " -0.044767\n", + " -0.080880\n", + " 0.062666\n", " \n", " \n", " 44\n", @@ -1275,16 +1328,16 @@ " 163840\n", " 227488\n", " ...\n", - " 0.002434\n", - " 0.001946\n", - " 0.001824\n", - " 0.000374\n", - " -0.000416\n", - " -0.000871\n", - " 0.003142\n", - " -1.690984e-03\n", - " 0.001557\n", - " 0.002596\n", + " -0.260044\n", + " -0.168338\n", + " -0.271208\n", + " -0.247191\n", + " -0.235198\n", + " -0.383056\n", + " -0.163678\n", + " -0.055487\n", + " -0.088683\n", + " -0.155701\n", " \n", " \n", " 45\n", @@ -1299,16 +1352,16 @@ " 163840\n", " 227488\n", " ...\n", - " -0.005434\n", - " 0.003420\n", - " -0.019319\n", - " -0.000747\n", - " 0.041849\n", - " 0.022034\n", - " -0.024061\n", - " 4.308212e-02\n", - " 0.005673\n", - " -0.007195\n", + " 0.107979\n", + " 0.020519\n", + " 0.216325\n", + " 0.572342\n", + " 0.147766\n", + " 0.106553\n", + " 0.096116\n", + " 0.279162\n", + " 0.162525\n", + " 0.332815\n", " \n", " \n", " 46\n", @@ -1323,16 +1376,16 @@ " 163840\n", " 174642\n", " ...\n", - " 0.000210\n", - " 0.000521\n", - " 0.000256\n", - " 0.000118\n", - " 0.000335\n", - " 0.000174\n", - " 0.000136\n", - " 8.502305e-04\n", - " 0.000165\n", - " 0.000544\n", + " -0.256797\n", + " -0.163505\n", + " -0.266977\n", + " -0.247593\n", + " -0.233061\n", + " -0.386286\n", + " -0.155122\n", + " -0.059363\n", + " -0.081943\n", + " -0.156882\n", " \n", " \n", " 47\n", @@ -1347,16 +1400,16 @@ " 163840\n", " 174642\n", " ...\n", - " -0.000382\n", - " -0.000466\n", - " -0.000408\n", - " -0.000091\n", - " -0.000512\n", - " -0.000240\n", - " -0.000283\n", - " -1.361221e-04\n", - " -0.000274\n", - " -0.000745\n", + " 0.100775\n", + " 0.018718\n", + " 0.213268\n", + " 0.574074\n", + " 0.151917\n", + " 0.110753\n", + " 0.088787\n", + " 0.286400\n", + " 0.165354\n", + " 0.329970\n", " \n", " \n", "\n", @@ -1415,127 +1468,495 @@ "47 chr1 109728807 TTT G ALX3 109710224 110234512 - \n", "\n", " gene_mask_start gene_mask_end ... agg_9528 agg_9529 agg_9530 \\\n", - "0 163840 172672 ... -0.000053 -0.000153 -0.000089 \n", - "1 163840 178946 ... -0.000586 -0.000891 -0.000565 \n", - "2 163840 170729 ... -0.000565 -0.000787 -0.000515 \n", - "3 163840 165050 ... -0.001453 -0.001775 -0.001403 \n", - "4 163840 166306 ... 0.000045 0.000105 0.000060 \n", - "5 163840 167179 ... -0.000561 -0.000826 -0.000536 \n", - "6 163840 198383 ... -0.000399 -0.000585 -0.000374 \n", - "7 163840 184493 ... 0.001317 0.001001 0.000884 \n", - "8 163840 168975 ... -0.000244 -0.000214 -0.000310 \n", - "9 163840 173223 ... 0.000012 -0.000113 0.000002 \n", - "10 163840 170339 ... -0.000286 -0.000486 -0.000312 \n", - "11 163840 177242 ... 0.007558 0.004934 0.007966 \n", - "12 163840 166924 ... -0.000190 -0.000248 -0.000215 \n", - "13 163840 166653 ... -0.000217 -0.000332 -0.000225 \n", - "14 163840 199838 ... -0.000643 -0.000886 -0.000513 \n", - "15 163840 178999 ... -0.000031 -0.000034 -0.000048 \n", - "16 163840 168116 ... -0.000396 -0.000607 -0.000421 \n", - "17 163840 183816 ... -0.000809 -0.001129 -0.000884 \n", - "18 163840 181059 ... -0.000826 -0.001292 -0.000932 \n", - "19 163840 176946 ... -0.000033 -0.000067 -0.000013 \n", - "20 163840 177982 ... -0.000297 -0.000456 -0.000343 \n", - "21 163840 172928 ... -0.000118 -0.000153 -0.000130 \n", - "22 163840 180515 ... 0.000099 0.000077 0.000057 \n", - "23 163840 180515 ... 0.003284 0.005105 0.002662 \n", - "24 163840 179428 ... 0.002200 0.001744 0.001206 \n", - "25 163840 179428 ... -0.004052 -0.003465 -0.003092 \n", - "26 163840 173165 ... 0.000001 0.000010 0.000091 \n", - "27 163840 173165 ... 0.000536 -0.000179 0.001585 \n", - "28 163840 172720 ... 0.000584 0.000577 0.000674 \n", - "29 163840 172720 ... 0.003762 0.002849 0.001487 \n", - "30 163840 172374 ... 0.000155 0.000251 0.000114 \n", - "31 163840 172374 ... 0.000258 0.000810 0.000308 \n", - "32 163840 170946 ... -0.000651 -0.000626 -0.000446 \n", - "33 163840 170946 ... 0.014822 0.029157 0.015316 \n", - "34 163840 233549 ... 0.002052 0.002637 0.003581 \n", - "35 163840 233549 ... -0.002514 -0.002176 0.001273 \n", - "36 163840 179789 ... 0.000870 0.001771 0.002003 \n", - "37 163840 179789 ... 0.015270 0.004848 0.011452 \n", - "38 163840 182577 ... -0.000336 0.001399 0.000511 \n", - "39 163840 182577 ... -0.032543 -0.033684 -0.029430 \n", - "40 163840 205369 ... -0.001145 -0.001371 0.000107 \n", - "41 163840 205369 ... 0.006410 0.010421 0.008430 \n", - "42 163840 185065 ... 0.000849 0.001993 0.001033 \n", - "43 163840 185065 ... 0.009973 0.012482 0.003073 \n", - "44 163840 227488 ... 0.002434 0.001946 0.001824 \n", - "45 163840 227488 ... -0.005434 0.003420 -0.019319 \n", - "46 163840 174642 ... 0.000210 0.000521 0.000256 \n", - "47 163840 174642 ... -0.000382 -0.000466 -0.000408 \n", + "0 163840 172672 ... -0.038816 -0.132628 0.012945 \n", + "1 163840 178946 ... 0.007422 -0.040294 0.037184 \n", + "2 163840 170729 ... -0.038910 -0.132768 0.012589 \n", + "3 163840 165050 ... 0.006907 -0.041188 0.036539 \n", + "4 163840 166306 ... 0.384163 0.383833 0.208923 \n", + "5 163840 167179 ... 0.128392 0.201155 0.096084 \n", + "6 163840 198383 ... 0.382839 0.387868 0.207169 \n", + "7 163840 184493 ... 0.127002 0.199668 0.092870 \n", + "8 163840 168975 ... -0.065035 -0.009808 -0.089304 \n", + "9 163840 173223 ... 0.149715 0.036652 0.236477 \n", + "10 163840 170339 ... -0.064135 -0.008839 -0.088697 \n", + "11 163840 177242 ... 0.149233 0.036425 0.236163 \n", + "12 163840 166924 ... 0.025078 0.316090 0.052550 \n", + "13 163840 166653 ... -0.060781 0.136466 0.222455 \n", + "14 163840 199838 ... 0.024326 0.315336 0.050626 \n", + "15 163840 178999 ... -0.069126 0.131456 0.215256 \n", + "16 163840 168116 ... -0.068721 -0.131200 -0.095311 \n", + "17 163840 183816 ... -0.000722 -0.065477 -0.015448 \n", + "18 163840 181059 ... -0.068234 -0.130586 -0.095205 \n", + "19 163840 176946 ... -0.000299 -0.064919 -0.015113 \n", + "20 163840 177982 ... 0.144722 0.096918 0.053467 \n", + "21 163840 172928 ... -0.204408 -0.164630 -0.161219 \n", + "22 163840 180515 ... 0.145138 0.099315 0.055320 \n", + "23 163840 180515 ... -0.216383 -0.174637 -0.171691 \n", + "24 163840 179428 ... 0.125082 0.187445 0.094098 \n", + "25 163840 179428 ... -0.072443 -0.081550 -0.038486 \n", + "26 163840 173165 ... 0.125406 0.187519 0.094387 \n", + "27 163840 173165 ... -0.072282 -0.081517 -0.038386 \n", + "28 163840 172720 ... -0.109378 -0.095939 -0.046820 \n", + "29 163840 172720 ... -0.014295 -0.047262 0.040566 \n", + "30 163840 172374 ... -0.114771 -0.101728 -0.051283 \n", + "31 163840 172374 ... -0.014773 -0.051328 0.039865 \n", + "32 163840 170946 ... 0.017597 -0.080504 0.049995 \n", + "33 163840 170946 ... -0.297318 0.036640 0.039409 \n", + "34 163840 233549 ... 0.017213 -0.081016 0.048742 \n", + "35 163840 233549 ... -0.296310 0.027849 0.034702 \n", + "36 163840 179789 ... 0.173036 0.238832 -0.112437 \n", + "37 163840 179789 ... -0.587394 -0.383097 -0.074972 \n", + "38 163840 182577 ... 0.196702 0.244239 -0.096241 \n", + "39 163840 182577 ... -0.598326 -0.350383 -0.062013 \n", + "40 163840 205369 ... -0.065015 -0.134536 -0.140023 \n", + "41 163840 205369 ... -0.138112 -0.135524 -0.122887 \n", + "42 163840 185065 ... -0.064936 -0.134381 -0.140410 \n", + "43 163840 185065 ... -0.138833 -0.136745 -0.123534 \n", + "44 163840 227488 ... -0.260044 -0.168338 -0.271208 \n", + "45 163840 227488 ... 0.107979 0.020519 0.216325 \n", + "46 163840 174642 ... -0.256797 -0.163505 -0.266977 \n", + "47 163840 174642 ... 0.100775 0.018718 0.213268 \n", "\n", - " agg_9531 agg_9532 agg_9533 agg_9535 agg_9536 agg_9537 agg_9538 \n", - "0 -0.000016 -0.000013 -0.000040 -0.000070 7.629395e-05 -0.000026 -0.000069 \n", - "1 -0.000311 -0.000461 -0.000252 -0.000376 -6.060898e-04 -0.000454 -0.000725 \n", - "2 -0.000279 -0.000354 -0.000202 -0.000342 -5.399883e-04 -0.000423 -0.000597 \n", - "3 -0.000820 -0.000896 -0.000575 -0.001113 -1.119316e-03 -0.001036 -0.001202 \n", - "4 -0.000030 -0.000049 0.000029 0.000135 1.490116e-07 0.000057 0.000001 \n", - "5 -0.000299 -0.000324 -0.000196 -0.000374 -3.589988e-04 -0.000342 -0.000426 \n", - "6 -0.000208 -0.000212 -0.000126 -0.000265 -2.926290e-04 -0.000266 -0.000376 \n", - "7 0.001013 0.000691 0.001310 0.001097 5.598068e-04 0.000501 0.000827 \n", - "8 -0.000153 -0.000324 -0.000168 -0.000147 -4.823208e-04 -0.000211 -0.000401 \n", - "9 -0.000028 -0.000248 -0.000283 -0.000314 -1.890063e-04 -0.000022 -0.000029 \n", - "10 -0.000150 -0.000256 -0.000124 -0.000179 -3.057718e-04 -0.000183 -0.000252 \n", - "11 0.007595 -0.001943 0.001598 0.007838 2.528191e-03 0.007137 0.009534 \n", - "12 -0.000113 -0.000070 -0.000076 -0.000159 -5.897880e-05 -0.000117 -0.000182 \n", - "13 -0.000132 -0.000127 -0.000102 -0.000146 -2.127588e-04 -0.000168 -0.000276 \n", - "14 -0.001688 -0.000346 -0.000528 -0.000894 -4.293919e-04 -0.001309 -0.000301 \n", - "15 -0.000017 0.000019 -0.000002 -0.000036 -1.719594e-05 -0.000026 -0.000037 \n", - "16 -0.000306 -0.000294 -0.000207 -0.000237 -3.867447e-04 -0.000289 -0.000498 \n", - "17 -0.000664 -0.000741 -0.000528 -0.000512 -7.908940e-04 -0.000607 -0.000990 \n", - "18 -0.000710 -0.000789 -0.000506 -0.000457 -9.316206e-04 -0.000666 -0.001122 \n", - "19 -0.000010 0.000030 0.000017 -0.000020 1.147389e-04 -0.000005 -0.000017 \n", - "20 -0.000253 -0.000310 -0.000204 -0.000172 -3.470182e-04 -0.000220 -0.000371 \n", - "21 -0.000090 -0.000128 -0.000076 -0.000053 -2.347827e-04 -0.000090 -0.000214 \n", - "22 0.000040 0.000018 0.000047 0.000075 2.750456e-04 0.000065 0.000258 \n", - "23 0.001423 0.001525 0.001257 0.001919 5.420089e-03 0.002166 0.005263 \n", - "24 0.001263 0.001008 -0.000117 0.001218 2.799034e-04 0.001503 0.001372 \n", - "25 -0.003744 -0.003384 -0.003265 -0.003253 1.600981e-03 -0.001915 -0.002728 \n", - "26 -0.000079 -0.000044 -0.000061 -0.000042 -4.082918e-05 -0.000022 -0.000060 \n", - "27 0.001431 0.003103 0.001631 0.000257 2.081454e-03 0.000913 0.001677 \n", - "28 0.000536 0.000167 0.000397 0.000507 1.705885e-04 0.000503 0.000481 \n", - "29 0.002733 -0.000099 0.000631 0.001469 4.820824e-04 0.002098 -0.000244 \n", - "30 0.000134 -0.000137 -0.000068 0.000125 4.252791e-05 0.000117 0.000137 \n", - "31 0.000489 0.000618 0.000236 0.000523 6.991625e-04 0.000155 0.000536 \n", - "32 -0.000290 -0.000174 -0.000026 -0.000154 -3.945976e-04 -0.000396 -0.000499 \n", - "33 0.007980 0.015211 0.007929 0.010251 2.181430e-02 0.009214 0.024373 \n", - "34 0.001084 -0.001867 -0.003804 0.003472 -1.005411e-03 0.003400 0.001397 \n", - "35 0.007376 0.011271 0.019305 -0.002886 2.035785e-02 -0.001660 0.006990 \n", - "36 0.000641 -0.002239 -0.000491 0.001610 -6.132126e-04 0.002603 0.000714 \n", - "37 0.028524 0.009958 0.013799 0.007597 2.641273e-02 0.009691 0.026180 \n", - "38 0.000450 -0.000501 -0.000243 -0.001328 4.479885e-04 0.000544 0.002848 \n", - "39 -0.050647 -0.030112 -0.021851 -0.012187 -2.968431e-02 -0.037312 -0.067244 \n", - "40 -0.000453 -0.001597 -0.000798 -0.001882 1.063347e-04 0.000106 0.000086 \n", - "41 0.010849 -0.004994 0.003090 -0.004843 -9.270430e-03 0.008755 0.011827 \n", - "42 -0.000050 -0.000534 0.000095 0.000409 -3.244877e-04 0.000842 0.000388 \n", - "43 0.019083 0.010606 0.000387 0.003512 7.757962e-03 0.011860 0.019928 \n", - "44 0.000374 -0.000416 -0.000871 0.003142 -1.690984e-03 0.001557 0.002596 \n", - "45 -0.000747 0.041849 0.022034 -0.024061 4.308212e-02 0.005673 -0.007195 \n", - "46 0.000118 0.000335 0.000174 0.000136 8.502305e-04 0.000165 0.000544 \n", - "47 -0.000091 -0.000512 -0.000240 -0.000283 -1.361221e-04 -0.000274 -0.000745 \n", + " agg_9531 agg_9532 agg_9533 agg_9535 agg_9536 agg_9537 agg_9538 \n", + "0 -0.057034 -0.045095 -0.019845 -0.048861 -0.248287 -0.157361 -0.202191 \n", + "1 -0.039102 -0.069273 -0.089758 0.085668 -0.100051 0.199948 -0.228135 \n", + "2 -0.057297 -0.044415 -0.019266 -0.049203 -0.247513 -0.158220 -0.202740 \n", + "3 -0.038984 -0.070001 -0.090065 0.085498 -0.100311 0.199196 -0.229159 \n", + "4 0.197689 0.147001 0.214184 0.429241 0.046892 0.232819 0.122424 \n", + "5 0.117298 -0.126285 -0.059694 0.184677 -0.009753 0.305165 -0.107976 \n", + "6 0.192711 0.148570 0.212126 0.429055 0.040387 0.231833 0.117729 \n", + "7 0.113858 -0.132513 -0.059923 0.180629 -0.008512 0.302857 -0.112850 \n", + "8 -0.123990 -0.237687 -0.145043 -0.027265 -0.172405 -0.209000 -0.169588 \n", + "9 -0.033082 -0.059123 -0.109544 -0.116550 -0.056748 0.238428 -0.066090 \n", + "10 -0.123552 -0.236969 -0.144438 -0.026115 -0.171792 -0.208494 -0.170012 \n", + "11 -0.033192 -0.059114 -0.109789 -0.116972 -0.057177 0.238166 -0.067322 \n", + "12 -0.128108 -0.025624 0.018586 0.070953 0.035494 0.035036 0.198598 \n", + "13 0.180580 -0.077465 0.059501 -0.152240 0.021208 0.388104 0.088033 \n", + "14 -0.121845 -0.024810 0.020802 0.065568 0.039766 0.038300 0.207067 \n", + "15 0.164175 -0.088568 0.049765 -0.160795 0.012626 0.377364 0.080307 \n", + "16 0.046062 0.058227 -0.014569 -0.110806 0.156508 0.050760 0.086102 \n", + "17 -0.048512 0.080478 0.044108 -0.040190 -0.109651 0.016822 -0.062620 \n", + "18 0.046179 0.059056 -0.013978 -0.110189 0.157281 0.050832 0.085974 \n", + "19 -0.048239 0.080818 0.044669 -0.040059 -0.109231 0.017035 -0.062525 \n", + "20 0.178367 0.186895 -0.086145 0.078297 0.339905 0.164013 0.351803 \n", + "21 -0.077524 0.001738 -0.079267 -0.283000 -0.061010 -0.083843 -0.061747 \n", + "22 0.177228 0.190383 -0.084195 0.083160 0.344412 0.166400 0.354686 \n", + "23 -0.086247 -0.017797 -0.098000 -0.294861 -0.081399 -0.096354 -0.078279 \n", + "24 0.073541 0.153524 0.122184 0.178186 0.325020 0.131711 0.251710 \n", + "25 -0.047714 0.226176 0.045261 -0.126760 0.371777 0.060598 -0.005164 \n", + "26 0.074029 0.154567 0.122893 0.178522 0.326261 0.131973 0.251846 \n", + "27 -0.047584 0.226377 0.045594 -0.126571 0.372053 0.060832 -0.004993 \n", + "28 -0.160233 -0.108172 -0.116325 0.049603 0.171065 0.111874 0.250584 \n", + "29 0.034287 0.086552 -0.002746 -0.036511 0.105897 0.075976 -0.094442 \n", + "30 -0.167664 -0.115162 -0.118221 0.045587 0.164287 0.105892 0.242462 \n", + "31 0.034642 0.079301 -0.005801 -0.037709 0.102587 0.075023 -0.093098 \n", + "32 -0.011808 -0.271627 -0.061634 -0.068092 -0.169892 -0.139209 -0.134266 \n", + "33 0.172793 0.131960 0.099102 -0.127269 0.248293 0.078142 0.244484 \n", + "34 -0.012799 -0.271882 -0.061404 -0.068229 -0.168541 -0.140811 -0.135465 \n", + "35 0.172035 0.133321 0.101695 -0.133212 0.247434 0.074210 0.243297 \n", + "36 -0.396939 -0.048346 -0.349238 0.178761 -0.541497 -0.432859 -0.115770 \n", + "37 0.116038 -0.106850 -0.536714 -0.289871 0.275144 -0.386951 -0.475129 \n", + "38 -0.383522 -0.053860 -0.346986 0.204916 -0.547306 -0.428087 -0.109396 \n", + "39 0.137351 -0.113740 -0.549226 -0.316263 0.282046 -0.367199 -0.457674 \n", + "40 0.024319 -0.060507 -0.108153 -0.149791 -0.038775 -0.076944 -0.118178 \n", + "41 0.003460 -0.095921 -0.142851 -0.133408 -0.044298 -0.079864 0.063350 \n", + "42 0.024382 -0.060211 -0.106950 -0.150407 -0.038308 -0.077518 -0.118339 \n", + "43 0.002528 -0.096342 -0.143056 -0.133687 -0.044767 -0.080880 0.062666 \n", + "44 -0.247191 -0.235198 -0.383056 -0.163678 -0.055487 -0.088683 -0.155701 \n", + "45 0.572342 0.147766 0.106553 0.096116 0.279162 0.162525 0.332815 \n", + "46 -0.247593 -0.233061 -0.386286 -0.155122 -0.059363 -0.081943 -0.156882 \n", + "47 0.574074 0.151917 0.110753 0.088787 0.286400 0.165354 0.329970 \n", "\n", "[48 rows x 8870 columns]" ] }, - "execution_count": 5, + "execution_count": 6, "metadata": {}, "output_type": "execute_result" } ], "source": [ - "pd.read_parquet(\"vep_vcf_results.parquet\")" + "results = pd.read_parquet(\"vep_vcf_results.parquet\")\n", + "results" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ - "Alternatively, you can pass tsv file with following format where first 4 columns are `chrom`, `pos`, `ref`, `alt`." + "Now, we need to match the track labels (agg_1, agg_2, etc.) to their metadata. You can load the metadata as follows:" ] }, { "cell_type": "code", - "execution_count": 6, + "execution_count": 7, + "metadata": {}, + "outputs": [ + { + "name": "stderr", + "output_type": "stream", + "text": [ + "/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n", + " from .autonotebook import tqdm as notebook_tqdm\n" + ] + }, + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
cell_typetissueorgandiseasestudydatasetregionsubregioncelltype_coarsen_cellstotal_countsn_genessize_factortrain_pearsonval_pearsontest_pearson
agg_0Amygdala excitatoryAmygdala_AmygdalaCNShealthyjhpce#tran2021brain_atlasAmygdalaAmygdalaNaN3311.592883e+071700041431.4651860.9424590.8413770.865640
agg_1Amygdala excitatoryAmygdala_Basolateral nuclear group (BLN) - lat...CNShealthySCR_016152brain_atlasAmygdalaBasolateral nuclear group (BLN) - lateral nucl...NaN113692.952133e+081808040765.3414810.9430980.8389360.861092
agg_2Amygdala excitatoryAmygdala_Bed nucleus of stria terminalis and n...CNShealthySCR_016152brain_atlasAmygdalaBed nucleus of stria terminalis and nearby - BNSTNaN1392.593231e+061541842556.3870200.9521700.8545440.866654
agg_3Amygdala excitatoryAmygdala_Central nuclear group - CENCNShealthySCR_016152brain_atlasAmygdalaCentral nuclear group - CENNaN38929.946371e+071795942884.6414300.9597440.8635850.881554
agg_4Amygdala excitatoryAmygdala_Corticomedial nuclear group (CMN) - a...CNShealthySCR_016152brain_atlasAmygdalaCorticomedial nuclear group (CMN) - anterior c...NaN29451.281619e+081788541816.7419330.9513650.8543040.868902
...................................................
agg_9533vascular associated smooth muscle cellupper lobe of right lunglungNAENCODEscimilaritynannanNaN213.483375e+04851535404.9117680.7352130.6656470.654491
agg_9535vascular associated smooth muscle cellurinary bladderurinaryhealthyGSE129845scimilaritynannanNaN248.498500e+04733726189.4157890.8098520.6900220.656160
agg_9536vascular associated smooth muscle celluterusuterusNAENCODEscimilaritynannanNaN2725.700762e+051476944938.4038670.9153290.8089410.839993
agg_9537vascular associated smooth muscle celluterusuterushealthye5f58829-1a66-40b5-a624-9046778e74f5scimilaritynannanNaN4721.089170e+071451430145.4221520.8523390.7176820.727469
agg_9538vascular associated smooth muscle cellvasculaturevasculaturehealthye5f58829-1a66-40b5-a624-9046778e74f5scimilaritynannanNaN18535.992697e+071676436464.2733710.9098550.7804130.796351
\n", + "

8856 rows × 16 columns

\n", + "
" + ], + "text/plain": [ + " cell_type \\\n", + "agg_0 Amygdala excitatory \n", + "agg_1 Amygdala excitatory \n", + "agg_2 Amygdala excitatory \n", + "agg_3 Amygdala excitatory \n", + "agg_4 Amygdala excitatory \n", + "... ... \n", + "agg_9533 vascular associated smooth muscle cell \n", + "agg_9535 vascular associated smooth muscle cell \n", + "agg_9536 vascular associated smooth muscle cell \n", + "agg_9537 vascular associated smooth muscle cell \n", + "agg_9538 vascular associated smooth muscle cell \n", + "\n", + " tissue organ \\\n", + "agg_0 Amygdala_Amygdala CNS \n", + "agg_1 Amygdala_Basolateral nuclear group (BLN) - lat... CNS \n", + "agg_2 Amygdala_Bed nucleus of stria terminalis and n... CNS \n", + "agg_3 Amygdala_Central nuclear group - CEN CNS \n", + "agg_4 Amygdala_Corticomedial nuclear group (CMN) - a... CNS \n", + "... ... ... \n", + "agg_9533 upper lobe of right lung lung \n", + "agg_9535 urinary bladder urinary \n", + "agg_9536 uterus uterus \n", + "agg_9537 uterus uterus \n", + "agg_9538 vasculature vasculature \n", + "\n", + " disease study dataset \\\n", + "agg_0 healthy jhpce#tran2021 brain_atlas \n", + "agg_1 healthy SCR_016152 brain_atlas \n", + "agg_2 healthy SCR_016152 brain_atlas \n", + "agg_3 healthy SCR_016152 brain_atlas \n", + "agg_4 healthy SCR_016152 brain_atlas \n", + "... ... ... ... \n", + "agg_9533 NA ENCODE scimilarity \n", + "agg_9535 healthy GSE129845 scimilarity \n", + "agg_9536 NA ENCODE scimilarity \n", + "agg_9537 healthy e5f58829-1a66-40b5-a624-9046778e74f5 scimilarity \n", + "agg_9538 healthy e5f58829-1a66-40b5-a624-9046778e74f5 scimilarity \n", + "\n", + " region subregion \\\n", + "agg_0 Amygdala Amygdala \n", + "agg_1 Amygdala Basolateral nuclear group (BLN) - lateral nucl... \n", + "agg_2 Amygdala Bed nucleus of stria terminalis and nearby - BNST \n", + "agg_3 Amygdala Central nuclear group - CEN \n", + "agg_4 Amygdala Corticomedial nuclear group (CMN) - anterior c... \n", + "... ... ... \n", + "agg_9533 nan nan \n", + "agg_9535 nan nan \n", + "agg_9536 nan nan \n", + "agg_9537 nan nan \n", + "agg_9538 nan nan \n", + "\n", + " celltype_coarse n_cells total_counts n_genes size_factor \\\n", + "agg_0 NaN 331 1.592883e+07 17000 41431.465186 \n", + "agg_1 NaN 11369 2.952133e+08 18080 40765.341481 \n", + "agg_2 NaN 139 2.593231e+06 15418 42556.387020 \n", + "agg_3 NaN 3892 9.946371e+07 17959 42884.641430 \n", + "agg_4 NaN 2945 1.281619e+08 17885 41816.741933 \n", + "... ... ... ... ... ... \n", + "agg_9533 NaN 21 3.483375e+04 8515 35404.911768 \n", + "agg_9535 NaN 24 8.498500e+04 7337 26189.415789 \n", + "agg_9536 NaN 272 5.700762e+05 14769 44938.403867 \n", + "agg_9537 NaN 472 1.089170e+07 14514 30145.422152 \n", + "agg_9538 NaN 1853 5.992697e+07 16764 36464.273371 \n", + "\n", + " train_pearson val_pearson test_pearson \n", + "agg_0 0.942459 0.841377 0.865640 \n", + "agg_1 0.943098 0.838936 0.861092 \n", + "agg_2 0.952170 0.854544 0.866654 \n", + "agg_3 0.959744 0.863585 0.881554 \n", + "agg_4 0.951365 0.854304 0.868902 \n", + "... ... ... ... \n", + "agg_9533 0.735213 0.665647 0.654491 \n", + "agg_9535 0.809852 0.690022 0.656160 \n", + "agg_9536 0.915329 0.808941 0.839993 \n", + "agg_9537 0.852339 0.717682 0.727469 \n", + "agg_9538 0.909855 0.780413 0.796351 \n", + "\n", + "[8856 rows x 16 columns]" + ] + }, + "execution_count": 7, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "from decima import DecimaResult\n", + "metadata = DecimaResult.load().cell_metadata\n", + "metadata" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "Instead of a vcf, you can also pass a tsv file with the following format where the first 4 columns are `chrom`, `pos`, `ref`, `alt`." + ] + }, + { + "cell_type": "code", + "execution_count": 8, "metadata": {}, "outputs": [ { @@ -1559,83 +1980,65 @@ "cell_type": "markdown", "metadata": {}, "source": [ - "You can only run predictions for the variants closer to tss than 100kbp anyway these are the ones likely to be most impactful on the gene expression." + "You can limit predictions to variant-gene pairs with a maximum distance (say 100kbp)." ] }, { "cell_type": "code", - "execution_count": 7, + "execution_count": 9, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ - "decima.vep - INFO - Using device: cuda and genome: hg38\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Currently logged in as: \u001b[33mcelikm5\u001b[0m (\u001b[33mcelikm5-genentech\u001b[0m) to \u001b[32mhttps://api.wandb.ai\u001b[0m. Use \u001b[1m`wandb login --relogin`\u001b[0m to force relogin\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact decima_metadata:latest, 628.05MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:0.5 (1276.1MB/s)\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact decima_rep0:latest, 2155.88MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:1.1 (1916.9MB/s)\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact human_state_dict_fold0:latest, 709.30MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:0.5 (1417.4MB/s)\n", - "Using default `ModelCheckpoint`. Consider installing `litmodels` package to enable `LitModelCheckpoint` for automatic upload to the Lightning model registry.\n", + "decima - INFO - Using device: 0 and genome: hg38\n", + "Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.\n", + "decima - INFO - Performing predictions on VariantDataset(33 variants from ['chr1'] between 598861 and 110072228 bp from TSS)\n", + "/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/lightning_fabric/plugins/environments/slurm.py:204: PossibleUserWarning: The `srun` command is available on your system but is not used. HINT: If your intention is to run Lightning on SLURM, prepend your python command with `srun` like so: srun python /gpfs/scratchfs01/site/u/lala8/conda/envs/decima/bin ...\n", "GPU available: True (cuda), used: True\n", "TPU available: False, using: 0 TPU cores\n", - "HPU available: False, using: 0 HPUs\n", - "/home/celikm5/miniforge3/envs/decima/lib/python3.11/site-packages/pytorch_lightning/trainer/connectors/logger_connector/logger_connector.py:76: UserWarning: Starting from v1.9.0, `tensorboardX` has been removed as a dependency of the `pytorch_lightning` package, due to potential conflicts with other packages in the ML ecosystem. For this reason, `logger=True` will use `CSVLogger` as the default logger, unless the `tensorboard` or `tensorboardX` packages are found. Please `pip install lightning[extra]` or one of them to enable TensorBoard support by default\n", + "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n", + "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n", "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n", - "/home/celikm5/miniforge3/envs/decima/lib/python3.11/site-packages/pytorch_lightning/trainer/connectors/data_connector.py:425: PossibleUserWarning: The 'predict_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=95` in the `DataLoader` to improve performance.\n", - "Predicting DataLoader 0: 100%|██████████████████| 66/66 [00:13<00:00, 4.88it/s]\n", - "decima.vep - INFO - Warnings:\n", - "decima.vep - INFO - allele_mismatch_with_reference_genome: 10 alleles out of 33 predictions mismatched with the genome file /home/celikm5/.local/share/genomes/hg38/hg38.fa.If this is not expected, please check if you are using the correct genome version.\n", + "\u001b[2KPredicting \u001b[35m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m 66/66 \u001b[2m0:01:16 • 0:00:00\u001b[0m \u001b[2;4m0.74it/s\u001b[0m it/s\u001b[0m it/s\u001b[0m \n", + "\u001b[?25hdecima - WARNING - Warnings:\n", + "decima - WARNING - allele_mismatch_with_reference_genome: 5 alleles out of 33 predictions mismatched with the genome file /home/lala8/.local/share/genomes/hg38/hg38.fa. If this is not expected, please check if you are using the correct genome version.\n", "\u001b[0m" ] } ], "source": [ - "! decima vep -v \"data/variants.tsv\" -o \"vep_results.parquet\" --max-distance 100_000 --max-distance-type \"tss\"" + "! decima vep -v \"data/variants.tsv\" -o \"vep_results.parquet\" --max-distance 100_000 --distance-type \"tss\"" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ - "If you have already have mapping genes and variant, you can use this mapping so predictions only will be conducted between this pairs." + "If you have already have a mapping between genes and variants, you can use this mapping so predictions will only will be computed between these pairs. Otherwise, the variant effect will be computed for all genes within Decima's distance window." ] }, { "cell_type": "code", - "execution_count": 8, + "execution_count": 10, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ - "decima.vep - INFO - Using device: cuda and genome: hg38\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Currently logged in as: \u001b[33mcelikm5\u001b[0m (\u001b[33mcelikm5-genentech\u001b[0m) to \u001b[32mhttps://api.wandb.ai\u001b[0m. Use \u001b[1m`wandb login --relogin`\u001b[0m to force relogin\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact decima_metadata:latest, 628.05MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:0.5 (1311.1MB/s)\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact decima_rep0:latest, 2155.88MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:1.1 (1900.4MB/s)\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact human_state_dict_fold0:latest, 709.30MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:0.5 (1475.9MB/s)\n", - "Using default `ModelCheckpoint`. Consider installing `litmodels` package to enable `LitModelCheckpoint` for automatic upload to the Lightning model registry.\n", + "decima - INFO - Using device: 0 and genome: hg38\n", + "Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.\n", + "decima - INFO - Performing predictions on VariantDataset(2 variants from ['chr1'] between 837298 and 1361586 bp from TSS)\n", + "/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/lightning_fabric/plugins/environments/slurm.py:204: PossibleUserWarning: The `srun` command is available on your system but is not used. HINT: If your intention is to run Lightning on SLURM, prepend your python command with `srun` like so: srun python /gpfs/scratchfs01/site/u/lala8/conda/envs/decima/bin ...\n", "GPU available: True (cuda), used: True\n", "TPU available: False, using: 0 TPU cores\n", - "HPU available: False, using: 0 HPUs\n", - "/home/celikm5/miniforge3/envs/decima/lib/python3.11/site-packages/pytorch_lightning/trainer/connectors/logger_connector/logger_connector.py:76: UserWarning: Starting from v1.9.0, `tensorboardX` has been removed as a dependency of the `pytorch_lightning` package, due to potential conflicts with other packages in the ML ecosystem. For this reason, `logger=True` will use `CSVLogger` as the default logger, unless the `tensorboard` or `tensorboardX` packages are found. Please `pip install lightning[extra]` or one of them to enable TensorBoard support by default\n", + "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n", + "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n", "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n", - "/home/celikm5/miniforge3/envs/decima/lib/python3.11/site-packages/pytorch_lightning/trainer/connectors/data_connector.py:425: PossibleUserWarning: The 'predict_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=95` in the `DataLoader` to improve performance.\n", - "Predicting DataLoader 0: 100%|████████████████████| 4/4 [00:01<00:00, 3.79it/s]\n", - "\u001b[0m" + "\u001b[2KPredicting \u001b[35m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m 4/4 \u001b[2m0:00:04 • 0:00:00\u001b[0m \u001b[2;4m0.74it/s\u001b[0m [2;4m0.95it/s\u001b[0m \n", + "\u001b[?25h\u001b[0m" ] } ], @@ -1645,7 +2048,7 @@ }, { "cell_type": "code", - "execution_count": 9, + "execution_count": 11, "metadata": {}, "outputs": [ { @@ -1706,16 +2109,16 @@ " 163840\n", " 177242\n", " ...\n", - " -0.000746\n", - " 0.002301\n", - " 0.005067\n", - " 0.000135\n", - " -0.000559\n", - " -0.003155\n", - " 0.006503\n", - " -0.001059\n", - " -0.000566\n", - " 0.000964\n", + " -0.757529\n", + " -0.096537\n", + " -0.507969\n", + " -1.093793\n", + " -0.619370\n", + " -1.054376\n", + " -0.035199\n", + " -0.046525\n", + " -0.040908\n", + " -0.129383\n", " \n", " \n", " 1\n", @@ -1730,16 +2133,16 @@ " 163840\n", " 177242\n", " ...\n", - " 0.007558\n", - " 0.004934\n", - " 0.007966\n", - " 0.007595\n", - " -0.001943\n", - " 0.001598\n", - " 0.007838\n", - " 0.002528\n", - " 0.007137\n", - " 0.009534\n", + " 0.961100\n", + " 0.593476\n", + " 1.353795\n", + " 2.282793\n", + " 0.917987\n", + " 0.981627\n", + " 0.890721\n", + " 1.287703\n", + " 0.982055\n", + " 1.072436\n", " \n", " \n", "\n", @@ -1752,17 +2155,17 @@ "1 chr1 1002308 T C ISG15 837298 1361586 + 163840 \n", "\n", " gene_mask_end ... agg_9528 agg_9529 agg_9530 agg_9531 agg_9532 \\\n", - "0 177242 ... -0.000746 0.002301 0.005067 0.000135 -0.000559 \n", - "1 177242 ... 0.007558 0.004934 0.007966 0.007595 -0.001943 \n", + "0 177242 ... -0.757529 -0.096537 -0.507969 -1.093793 -0.619370 \n", + "1 177242 ... 0.961100 0.593476 1.353795 2.282793 0.917987 \n", "\n", " agg_9533 agg_9535 agg_9536 agg_9537 agg_9538 \n", - "0 -0.003155 0.006503 -0.001059 -0.000566 0.000964 \n", - "1 0.001598 0.007838 0.002528 0.007137 0.009534 \n", + "0 -1.054376 -0.035199 -0.046525 -0.040908 -0.129383 \n", + "1 0.981627 0.890721 1.287703 0.982055 1.072436 \n", "\n", "[2 rows x 8870 columns]" ] }, - "execution_count": 9, + "execution_count": 11, "metadata": {}, "output_type": "execute_result" } @@ -1775,77 +2178,46 @@ "cell_type": "markdown", "metadata": {}, "source": [ - "The vep api reads n (default=10_000) number of variants from vcf file performs predictions on these variants, saves them to parquet file then performs predictios for next next chuck. You can change chucksize:" + "The vep api reads n (default=10_000) number of variants from vcf file, performs predictions on these variants, saves them to a parquet file, then performs predictions for the next chunk. You can change the chunksize:" ] }, { "cell_type": "code", - "execution_count": 10, + "execution_count": 12, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ - "decima.vep - INFO - Using device: cuda and genome: hg38\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Currently logged in as: \u001b[33mcelikm5\u001b[0m (\u001b[33mcelikm5-genentech\u001b[0m) to \u001b[32mhttps://api.wandb.ai\u001b[0m. Use \u001b[1m`wandb login --relogin`\u001b[0m to force relogin\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact decima_metadata:latest, 628.05MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:0.5 (1365.3MB/s)\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact decima_rep0:latest, 2155.88MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:1.2 (1851.4MB/s)\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact human_state_dict_fold0:latest, 709.30MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:0.5 (1323.7MB/s)\n", - "Using default `ModelCheckpoint`. Consider installing `litmodels` package to enable `LitModelCheckpoint` for automatic upload to the Lightning model registry.\n", + "decima - INFO - Using device: 0 and genome: hg38\n", + "Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.\n", + "decima - INFO - Performing predictions on VariantDataset(22 variants from ['chr1'] between 516455 and 1525617 bp from TSS)\n", + "/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/lightning_fabric/plugins/environments/slurm.py:204: PossibleUserWarning: The `srun` command is available on your system but is not used. HINT: If your intention is to run Lightning on SLURM, prepend your python command with `srun` like so: srun python /gpfs/scratchfs01/site/u/lala8/conda/envs/decima/bin ...\n", "GPU available: True (cuda), used: True\n", "TPU available: False, using: 0 TPU cores\n", - "HPU available: False, using: 0 HPUs\n", - "/home/celikm5/miniforge3/envs/decima/lib/python3.11/site-packages/pytorch_lightning/trainer/connectors/logger_connector/logger_connector.py:76: UserWarning: Starting from v1.9.0, `tensorboardX` has been removed as a dependency of the `pytorch_lightning` package, due to potential conflicts with other packages in the ML ecosystem. For this reason, `logger=True` will use `CSVLogger` as the default logger, unless the `tensorboard` or `tensorboardX` packages are found. Please `pip install lightning[extra]` or one of them to enable TensorBoard support by default\n", + "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n", + "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n", "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n", - "/home/celikm5/miniforge3/envs/decima/lib/python3.11/site-packages/pytorch_lightning/trainer/connectors/data_connector.py:425: PossibleUserWarning: The 'predict_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=95` in the `DataLoader` to improve performance.\n", - "Predicting DataLoader 0: 100%|██████████████████| 44/44 [00:09<00:00, 4.86it/s]\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact decima_metadata:latest, 628.05MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:0.5 (1340.8MB/s)\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact decima_rep0:latest, 2155.88MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:1.1 (1944.0MB/s)\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact human_state_dict_fold0:latest, 709.30MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:0.4 (1576.5MB/s)\n", - "Using default `ModelCheckpoint`. Consider installing `litmodels` package to enable `LitModelCheckpoint` for automatic upload to the Lightning model registry.\n", + "\u001b[2KPredicting \u001b[35m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m 44/44 \u001b[2m0:00:44 • 0:00:00\u001b[0m \u001b[2;4m1.00it/s\u001b[0m it/s\u001b[0m it/s\u001b[0m \n", + "\u001b[?25hdecima - INFO - Performing predictions on VariantDataset(13 variants from ['chr1'] between 109259481 and 110234512 bp from TSS)\n", + "/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/lightning_fabric/plugins/environments/slurm.py:204: PossibleUserWarning: The `srun` command is available on your system but is not used. HINT: If your intention is to run Lightning on SLURM, prepend your python command with `srun` like so: srun python /gpfs/scratchfs01/site/u/lala8/conda/envs/decima/bin ...\n", "GPU available: True (cuda), used: True\n", "TPU available: False, using: 0 TPU cores\n", - "HPU available: False, using: 0 HPUs\n", + "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n", + "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n", "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n", - "/home/celikm5/miniforge3/envs/decima/lib/python3.11/site-packages/pytorch_lightning/trainer/connectors/data_connector.py:425: PossibleUserWarning: The 'predict_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=95` in the `DataLoader` to improve performance.\n", - "Predicting DataLoader 0: 100%|██████████████████| 26/26 [00:05<00:00, 4.98it/s]\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact decima_metadata:latest, 628.05MB. 1 files... \n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:0.5 (1359.5MB/s)\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact decima_rep0:latest, 2155.88MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:1.1 (1939.7MB/s)\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact human_state_dict_fold0:latest, 709.30MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:0.5 (1425.5MB/s)\n", - "Using default `ModelCheckpoint`. Consider installing `litmodels` package to enable `LitModelCheckpoint` for automatic upload to the Lightning model registry.\n", + "\u001b[2KPredicting \u001b[35m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m 26/26 \u001b[2m0:00:25 • 0:00:00\u001b[0m \u001b[2;4m0.96it/s\u001b[0m [2;4m1.00it/s\u001b[0m \n", + "\u001b[?25hdecima - INFO - Performing predictions on VariantDataset(13 variants from ['chr1'] between 109259481 and 110234512 bp from TSS)\n", + "/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/lightning_fabric/plugins/environments/slurm.py:204: PossibleUserWarning: The `srun` command is available on your system but is not used. HINT: If your intention is to run Lightning on SLURM, prepend your python command with `srun` like so: srun python /gpfs/scratchfs01/site/u/lala8/conda/envs/decima/bin ...\n", "GPU available: True (cuda), used: True\n", "TPU available: False, using: 0 TPU cores\n", - "HPU available: False, using: 0 HPUs\n", + "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n", + "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n", "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n", - "/home/celikm5/miniforge3/envs/decima/lib/python3.11/site-packages/pytorch_lightning/trainer/connectors/data_connector.py:425: PossibleUserWarning: The 'predict_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=95` in the `DataLoader` to improve performance.\n", - "Predicting DataLoader 0: 100%|██████████████████| 26/26 [00:05<00:00, 4.97it/s]\n", - "decima.vep - INFO - Warnings:\n", - "decima.vep - INFO - allele_mismatch_with_reference_genome: 26 alleles out of 48 predictions mismatched with the genome file /home/celikm5/.local/share/genomes/hg38/hg38.fa.If this is not expected, please check if you are using the correct genome version.\n", + "\u001b[2KPredicting \u001b[35m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m 26/26 \u001b[2m0:00:45 • 0:00:00\u001b[0m \u001b[2;4m0.57it/s\u001b[0m [2;4m0.54it/s\u001b[0m \n", + "\u001b[?25hdecima - WARNING - Warnings:\n", + "decima - WARNING - allele_mismatch_with_reference_genome: 10 alleles out of 48 predictions mismatched with the genome file /home/lala8/.local/share/genomes/hg38/hg38.fa. If this is not expected, please check if you are using the correct genome version.\n", "\u001b[0m" ] } @@ -1870,7 +2242,7 @@ }, { "cell_type": "code", - "execution_count": 11, + "execution_count": 13, "metadata": {}, "outputs": [], "source": [ @@ -1885,7 +2257,7 @@ }, { "cell_type": "code", - "execution_count": 12, + "execution_count": 14, "metadata": {}, "outputs": [ { @@ -1964,7 +2336,7 @@ "4 chr1 109728807 T GG" ] }, - "execution_count": 12, + "execution_count": 14, "metadata": {}, "output_type": "execute_result" } @@ -1978,50 +2350,121 @@ "cell_type": "markdown", "metadata": {}, "source": [ - "Simply pass your dataframe to `predict_variat_effect` function which will return dataframe for the prediction. You can pass `tasks` query to subset predictions for specific cells. Moreover, by default decima model for replicate 0 is used to use other replicates pass model=`1` , `2` or `3` to use other replicates or pass your custom model. If you pass `include_cols` argument the columns in the input will maintained in the output. To further variants based on distance to tss use `max_dist_tss` argument." + "Simply pass your dataframe to `predict_variant_effect` function which will return dataframe for the prediction. You can pass `tasks` query to subset predictions for specific cells. Moreover, by default the ensemble of 4 replicates is used. To use a specific replicate, pass model= `0`, `1` , `2` or `3` or pass your custom model. If you pass `include_cols` argument the columns in the input will be maintained in the output. To further select variants based on distance to tss use the `max_distance` argument." ] }, { "cell_type": "code", - "execution_count": 13, + "execution_count": 15, + "metadata": {}, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "\u001b[31mSignature:\u001b[39m\n", + "predict_variant_effect(\n", + " df_variant: Union[pandas.core.frame.DataFrame, str],\n", + " output_pq: Optional[str] = \u001b[38;5;28;01mNone\u001b[39;00m,\n", + " tasks: Union[str, List[str], NoneType] = \u001b[38;5;28;01mNone\u001b[39;00m,\n", + " model: Union[int, str, List[str]] = \u001b[33m'ensemble'\u001b[39m,\n", + " metadata_anndata: Optional[str] = \u001b[38;5;28;01mNone\u001b[39;00m,\n", + " chunksize: int = \u001b[32m10000\u001b[39m,\n", + " batch_size: int = \u001b[32m1\u001b[39m,\n", + " num_workers: int = \u001b[32m16\u001b[39m,\n", + " device: Optional[str] = \u001b[38;5;28;01mNone\u001b[39;00m,\n", + " include_cols: Optional[List[str]] = \u001b[38;5;28;01mNone\u001b[39;00m,\n", + " gene_col: Optional[str] = \u001b[38;5;28;01mNone\u001b[39;00m,\n", + " distance_type: Optional[str] = \u001b[33m'tss'\u001b[39m,\n", + " min_distance: Optional[float] = \u001b[32m0\u001b[39m,\n", + " max_distance: Optional[float] = inf,\n", + " genome: str = \u001b[33m'hg38'\u001b[39m,\n", + " save_replicates: bool = \u001b[38;5;28;01mFalse\u001b[39;00m,\n", + " reference_cache: bool = \u001b[38;5;28;01mTrue\u001b[39;00m,\n", + " float_precision: str = \u001b[33m'32'\u001b[39m,\n", + ") -> \u001b[38;5;28;01mNone\u001b[39;00m\n", + "\u001b[31mDocstring:\u001b[39m\n", + "Predict variant effect and save to parquet\n", + "\n", + "Args:\n", + " df_variant (pd.DataFrame or str): DataFrame with variant information or path to variant file\n", + " output_pq (str, optional): Path to save the parquet file. Defaults to None.\n", + " tasks (str, optional): Tasks to predict. Defaults to None.\n", + " model (int, optional): Model to use. Defaults to DEFAULT_ENSEMBLE.\n", + " metadata_anndata (str, optional): Path to anndata file. Defaults to None.\n", + " chunksize (int, optional): Number of variants to predict in each chunk. Defaults to 10_000.\n", + " batch_size (int, optional): Batch size. Defaults to 1.\n", + " num_workers (int, optional): Number of workers. Defaults to 16.\n", + " device (str, optional): Device to use. Defaults to None.\n", + " include_cols (list, optional): Columns to include in the output. Defaults to None.\n", + " gene_col (str, optional): Column name for gene names. Defaults to None.\n", + " distance_type (str, optional): Type of distance. Defaults to \"tss\".\n", + " min_distance (float, optional): Minimum distance from the end of the gene. Defaults to 0 (inclusive).\n", + " max_distance (float, optional): Maximum distance from the TSS. Defaults to inf (exclusive).\n", + " genome (str, optional): Genome name or path to the genome fasta file. Defaults to \"hg38\".\n", + " save_replicates (bool, optional): Save the replicates in the output. Defaults to False.\n", + " reference_cache (bool, optional): Whether to use reference cache. Defaults to True.\n", + " float_precision (str, optional): Floating-point precision. Defaults to \"32\".\n", + "\u001b[31mFile:\u001b[39m ~/decima/src/decima/vep/vep.py\n", + "\u001b[31mType:\u001b[39m function" + ] + } + ], + "source": [ + "predict_variant_effect?" + ] + }, + { + "cell_type": "code", + "execution_count": 16, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Currently logged in as: \u001b[33mcelikm5\u001b[0m (\u001b[33mcelikm5-genentech\u001b[0m) to \u001b[32mhttps://api.wandb.ai\u001b[0m. Use \u001b[1m`wandb login --relogin`\u001b[0m to force relogin\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact decima_metadata:latest, 628.05MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:0.5 (1339.8MB/s)\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact decima_rep0:latest, 2155.88MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:1.2 (1791.0MB/s)\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact human_state_dict_fold0:latest, 709.30MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:0.5 (1495.3MB/s)\n", - "Using default `ModelCheckpoint`. Consider installing `litmodels` package to enable `LitModelCheckpoint` for automatic upload to the Lightning model registry.\n", + "Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.\n", + "/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/lightning_fabric/plugins/environments/slurm.py:204: PossibleUserWarning: The `srun` command is available on your system but is not used. HINT: If your intention is to run Lightning on SLURM, prepend your python command with `srun` like so: srun python /gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib ...\n", "GPU available: True (cuda), used: True\n", "TPU available: False, using: 0 TPU cores\n", - "HPU available: False, using: 0 HPUs\n", - "/home/celikm5/miniforge3/envs/decima/lib/python3.11/site-packages/pytorch_lightning/trainer/connectors/logger_connector/logger_connector.py:76: UserWarning: Starting from v1.9.0, `tensorboardX` has been removed as a dependency of the `pytorch_lightning` package, due to potential conflicts with other packages in the ML ecosystem. For this reason, `logger=True` will use `CSVLogger` as the default logger, unless the `tensorboard` or `tensorboardX` packages are found. Please `pip install lightning[extra]` or one of them to enable TensorBoard support by default\n", + "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n", + "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n", "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n" ] }, { "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "a73a894cc53a4b2f98c395e3b8c6bf72", - "version_major": 2, - "version_minor": 0 - }, + "text/html": [ + "
/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/rich/live.py:260: UserWarning: \n",
+       "install \"ipywidgets\" for Jupyter support\n",
+       "
\n" + ], "text/plain": [ - "Predicting: | | 0/? [00:00\n" + ], + "text/plain": [] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stderr", + "output_type": "stream", + "text": [ + "Warnings:\n", + "allele_mismatch_with_reference_genome: 13 alleles out of 82 predictions mismatched with the genome file /home/lala8/.local/share/genomes/hg38/hg38.fa. If this is not expected, please check if you are using the correct genome version.\n" + ] + }, { "data": { "text/html": [ @@ -2080,16 +2523,16 @@ " 163840\n", " 172672\n", " ...\n", - " -0.003487\n", - " -0.006149\n", - " -0.003175\n", - " -0.002726\n", - " -0.003147\n", - " -0.001746\n", - " -0.002291\n", - " -0.005283\n", - " -0.003393\n", - " -0.006694\n", + " -0.006802\n", + " -0.043372\n", + " -0.031292\n", + " -0.002179\n", + " 0.046106\n", + " -0.025365\n", + " -0.086671\n", + " 0.013381\n", + " -0.037975\n", + " -0.062525\n", " \n", " \n", " 1\n", @@ -2104,16 +2547,16 @@ " 163840\n", " 172672\n", " ...\n", - " -0.000096\n", - " -0.000190\n", - " -0.000135\n", - " -0.000043\n", - " -0.000076\n", - " -0.000072\n", - " -0.000106\n", - " 0.000029\n", - " -0.000055\n", - " -0.000119\n", + " -0.245593\n", + " -0.190111\n", + " -0.137129\n", + " 0.051578\n", + " 0.003079\n", + " -0.128517\n", + " -0.236761\n", + " 0.017629\n", + " -0.094391\n", + " -0.070108\n", " \n", " \n", " 2\n", @@ -2128,16 +2571,16 @@ " 163840\n", " 178946\n", " ...\n", - " -0.002595\n", - " -0.004256\n", - " -0.002759\n", - " -0.001601\n", - " -0.002601\n", - " -0.001238\n", - " -0.001756\n", - " -0.002918\n", - " -0.001914\n", - " -0.003244\n", + " -0.005441\n", + " -0.041043\n", + " -0.028593\n", + " 0.001764\n", + " 0.048425\n", + " -0.024294\n", + " -0.086048\n", + " 0.018201\n", + " -0.035142\n", + " -0.059727\n", " \n", " \n", " 3\n", @@ -2152,16 +2595,16 @@ " 163840\n", " 178946\n", " ...\n", - " -0.000587\n", - " -0.000894\n", - " -0.000563\n", - " -0.000324\n", - " -0.000471\n", - " -0.000258\n", - " -0.000382\n", - " -0.000683\n", - " -0.000459\n", - " -0.000757\n", + " -0.241628\n", + " -0.187019\n", + " -0.134874\n", + " 0.050618\n", + " 0.005003\n", + " -0.125120\n", + " -0.233175\n", + " 0.019390\n", + " -0.093390\n", + " -0.068094\n", " \n", " \n", " 4\n", @@ -2176,16 +2619,16 @@ " 163840\n", " 170729\n", " ...\n", - " -0.002933\n", - " -0.004738\n", - " -0.003541\n", - " -0.001943\n", - " -0.002958\n", - " -0.001370\n", - " -0.002272\n", - " -0.003326\n", - " -0.002195\n", - " -0.003648\n", + " 0.152879\n", + " 0.291377\n", + " -0.198813\n", + " -0.429966\n", + " 0.270807\n", + " -0.239578\n", + " 0.068387\n", + " -0.406068\n", + " -0.293937\n", + " -0.101793\n", " \n", " \n", " ...\n", @@ -2224,16 +2667,16 @@ " 163840\n", " 227488\n", " ...\n", - " -0.004391\n", - " 0.004506\n", - " -0.017141\n", - " 0.000025\n", - " 0.042810\n", - " 0.024843\n", - " -0.022073\n", - " 0.043973\n", - " 0.005934\n", - " -0.006544\n", + " -0.391025\n", + " -0.056139\n", + " -0.131877\n", + " 0.002086\n", + " 0.052237\n", + " -0.073272\n", + " -0.171256\n", + " 0.151294\n", + " -0.122621\n", + " 0.059819\n", " \n", " \n", " 78\n", @@ -2248,16 +2691,16 @@ " 163840\n", " 227488\n", " ...\n", - " -0.129386\n", - " -0.144098\n", - " -0.077484\n", - " -0.091391\n", - " -0.104009\n", - " -0.034113\n", - " -0.096020\n", - " -0.063028\n", - " -0.079460\n", - " -0.102111\n", + " -0.026493\n", + " -0.118427\n", + " 0.084949\n", + " -0.022378\n", + " -0.210016\n", + " -0.091448\n", + " 0.038883\n", + " -0.119953\n", + " -0.181654\n", + " -0.168288\n", " \n", " \n", " 79\n", @@ -2272,16 +2715,16 @@ " 163840\n", " 174642\n", " ...\n", - " 0.000218\n", - " 0.000532\n", - " 0.000263\n", - " 0.000122\n", - " 0.000352\n", - " 0.000184\n", - " 0.000139\n", - " 0.000887\n", - " 0.000173\n", - " 0.000558\n", + " -0.414233\n", + " -0.028989\n", + " -0.079399\n", + " 0.069994\n", + " 0.022942\n", + " -0.047608\n", + " -0.206531\n", + " 0.133008\n", + " -0.047079\n", + " -0.011383\n", " \n", " \n", " 80\n", @@ -2296,16 +2739,16 @@ " 163840\n", " 174642\n", " ...\n", - " -0.000218\n", - " -0.000109\n", - " -0.000204\n", - " 0.000009\n", - " -0.000265\n", - " -0.000104\n", - " -0.000189\n", - " 0.000523\n", - " -0.000149\n", - " -0.000385\n", + " -0.084961\n", + " -0.115746\n", + " -0.085654\n", + " -0.038717\n", + " 0.006230\n", + " -0.039447\n", + " -0.157264\n", + " -0.002997\n", + " -0.075576\n", + " -0.172018\n", " \n", " \n", " 81\n", @@ -2320,16 +2763,16 @@ " 163840\n", " 174642\n", " ...\n", - " -0.001127\n", - " -0.002115\n", - " -0.001278\n", - " -0.000581\n", - " -0.001344\n", - " -0.000724\n", - " -0.000569\n", - " -0.003553\n", - " -0.000857\n", - " -0.002281\n", + " -0.223092\n", + " -0.099597\n", + " -0.119387\n", + " 0.070661\n", + " -0.016427\n", + " -0.110365\n", + " -0.208193\n", + " 0.088656\n", + " -0.018158\n", + " 0.014191\n", " \n", " \n", "\n", @@ -2351,35 +2794,35 @@ "81 chr1 109728807 T GG ALX3 109710224 110234512 - \n", "\n", " gene_mask_start gene_mask_end ... agg_9528 agg_9529 agg_9530 \\\n", - "0 163840 172672 ... -0.003487 -0.006149 -0.003175 \n", - "1 163840 172672 ... -0.000096 -0.000190 -0.000135 \n", - "2 163840 178946 ... -0.002595 -0.004256 -0.002759 \n", - "3 163840 178946 ... -0.000587 -0.000894 -0.000563 \n", - "4 163840 170729 ... -0.002933 -0.004738 -0.003541 \n", + "0 163840 172672 ... -0.006802 -0.043372 -0.031292 \n", + "1 163840 172672 ... -0.245593 -0.190111 -0.137129 \n", + "2 163840 178946 ... -0.005441 -0.041043 -0.028593 \n", + "3 163840 178946 ... -0.241628 -0.187019 -0.134874 \n", + "4 163840 170729 ... 0.152879 0.291377 -0.198813 \n", ".. ... ... ... ... ... ... \n", - "77 163840 227488 ... -0.004391 0.004506 -0.017141 \n", - "78 163840 227488 ... -0.129386 -0.144098 -0.077484 \n", - "79 163840 174642 ... 0.000218 0.000532 0.000263 \n", - "80 163840 174642 ... -0.000218 -0.000109 -0.000204 \n", - "81 163840 174642 ... -0.001127 -0.002115 -0.001278 \n", + "77 163840 227488 ... -0.391025 -0.056139 -0.131877 \n", + "78 163840 227488 ... -0.026493 -0.118427 0.084949 \n", + "79 163840 174642 ... -0.414233 -0.028989 -0.079399 \n", + "80 163840 174642 ... -0.084961 -0.115746 -0.085654 \n", + "81 163840 174642 ... -0.223092 -0.099597 -0.119387 \n", "\n", " agg_9531 agg_9532 agg_9533 agg_9535 agg_9536 agg_9537 agg_9538 \n", - "0 -0.002726 -0.003147 -0.001746 -0.002291 -0.005283 -0.003393 -0.006694 \n", - "1 -0.000043 -0.000076 -0.000072 -0.000106 0.000029 -0.000055 -0.000119 \n", - "2 -0.001601 -0.002601 -0.001238 -0.001756 -0.002918 -0.001914 -0.003244 \n", - "3 -0.000324 -0.000471 -0.000258 -0.000382 -0.000683 -0.000459 -0.000757 \n", - "4 -0.001943 -0.002958 -0.001370 -0.002272 -0.003326 -0.002195 -0.003648 \n", + "0 -0.002179 0.046106 -0.025365 -0.086671 0.013381 -0.037975 -0.062525 \n", + "1 0.051578 0.003079 -0.128517 -0.236761 0.017629 -0.094391 -0.070108 \n", + "2 0.001764 0.048425 -0.024294 -0.086048 0.018201 -0.035142 -0.059727 \n", + "3 0.050618 0.005003 -0.125120 -0.233175 0.019390 -0.093390 -0.068094 \n", + "4 -0.429966 0.270807 -0.239578 0.068387 -0.406068 -0.293937 -0.101793 \n", ".. ... ... ... ... ... ... ... \n", - "77 0.000025 0.042810 0.024843 -0.022073 0.043973 0.005934 -0.006544 \n", - "78 -0.091391 -0.104009 -0.034113 -0.096020 -0.063028 -0.079460 -0.102111 \n", - "79 0.000122 0.000352 0.000184 0.000139 0.000887 0.000173 0.000558 \n", - "80 0.000009 -0.000265 -0.000104 -0.000189 0.000523 -0.000149 -0.000385 \n", - "81 -0.000581 -0.001344 -0.000724 -0.000569 -0.003553 -0.000857 -0.002281 \n", + "77 0.002086 0.052237 -0.073272 -0.171256 0.151294 -0.122621 0.059819 \n", + "78 -0.022378 -0.210016 -0.091448 0.038883 -0.119953 -0.181654 -0.168288 \n", + "79 0.069994 0.022942 -0.047608 -0.206531 0.133008 -0.047079 -0.011383 \n", + "80 -0.038717 0.006230 -0.039447 -0.157264 -0.002997 -0.075576 -0.172018 \n", + "81 0.070661 -0.016427 -0.110365 -0.208193 0.088656 -0.018158 0.014191 \n", "\n", "[82 rows x 8870 columns]" ] }, - "execution_count": 13, + "execution_count": 16, "metadata": {}, "output_type": "execute_result" } @@ -2397,42 +2840,53 @@ }, { "cell_type": "code", - "execution_count": 14, + "execution_count": 17, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact decima_metadata:latest, 628.05MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:0.5 (1289.8MB/s)\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact decima_rep0:latest, 2155.88MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:1.2 (1774.2MB/s)\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact human_state_dict_fold0:latest, 709.30MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:0.5 (1430.2MB/s)\n", - "Using default `ModelCheckpoint`. Consider installing `litmodels` package to enable `LitModelCheckpoint` for automatic upload to the Lightning model registry.\n", + "/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/lightning_fabric/plugins/environments/slurm.py:204: PossibleUserWarning: The `srun` command is available on your system but is not used. HINT: If your intention is to run Lightning on SLURM, prepend your python command with `srun` like so: srun python /gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib ...\n", "GPU available: True (cuda), used: True\n", "TPU available: False, using: 0 TPU cores\n", - "HPU available: False, using: 0 HPUs\n", + "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n", + "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n", "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n" ] }, { "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "f9c815b42b1b4ba6ad1edb5427a53db6", - "version_major": 2, - "version_minor": 0 - }, + "text/html": [ + "
/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/rich/live.py:260: UserWarning: \n",
+       "install \"ipywidgets\" for Jupyter support\n",
+       "
\n" + ], "text/plain": [ - "Predicting: | | 0/? [00:00\n" + ], + "text/plain": [] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stderr", + "output_type": "stream", + "text": [ + "Warnings:\n", + "allele_mismatch_with_reference_genome: 13 alleles out of 82 predictions mismatched with the genome file /home/lala8/.local/share/genomes/hg38/hg38.fa. If this is not expected, please check if you are using the correct genome version.\n" + ] } ], "source": [ @@ -2441,7 +2895,7 @@ }, { "cell_type": "code", - "execution_count": 15, + "execution_count": 18, "metadata": {}, "outputs": [ { @@ -2502,16 +2956,16 @@ " 163840\n", " 172672\n", " ...\n", - " -0.003487\n", - " -0.006149\n", - " -0.003175\n", - " -0.002726\n", - " -0.003147\n", - " -0.001746\n", - " -0.002291\n", - " -0.005283\n", - " -0.003393\n", - " -0.006694\n", + " -0.006802\n", + " -0.043372\n", + " -0.031292\n", + " -0.002179\n", + " 0.046106\n", + " -0.025365\n", + " -0.086671\n", + " 0.013381\n", + " -0.037975\n", + " -0.062525\n", " \n", " \n", " 1\n", @@ -2526,16 +2980,16 @@ " 163840\n", " 172672\n", " ...\n", - " -0.000096\n", - " -0.000190\n", - " -0.000135\n", - " -0.000043\n", - " -0.000076\n", - " -0.000072\n", - " -0.000106\n", - " 0.000029\n", - " -0.000055\n", - " -0.000119\n", + " -0.245593\n", + " -0.190111\n", + " -0.137129\n", + " 0.051578\n", + " 0.003079\n", + " -0.128517\n", + " -0.236761\n", + " 0.017629\n", + " -0.094391\n", + " -0.070108\n", " \n", " \n", " 2\n", @@ -2550,16 +3004,16 @@ " 163840\n", " 178946\n", " ...\n", - " -0.002595\n", - " -0.004256\n", - " -0.002759\n", - " -0.001601\n", - " -0.002601\n", - " -0.001238\n", - " -0.001756\n", - " -0.002918\n", - " -0.001914\n", - " -0.003244\n", + " -0.005441\n", + " -0.041043\n", + " -0.028593\n", + " 0.001764\n", + " 0.048425\n", + " -0.024294\n", + " -0.086048\n", + " 0.018201\n", + " -0.035142\n", + " -0.059727\n", " \n", " \n", " 3\n", @@ -2574,16 +3028,16 @@ " 163840\n", " 178946\n", " ...\n", - " -0.000587\n", - " -0.000894\n", - " -0.000563\n", - " -0.000324\n", - " -0.000471\n", - " -0.000258\n", - " -0.000382\n", - " -0.000683\n", - " -0.000459\n", - " -0.000757\n", + " -0.241628\n", + " -0.187019\n", + " -0.134874\n", + " 0.050618\n", + " 0.005003\n", + " -0.125120\n", + " -0.233175\n", + " 0.019390\n", + " -0.093390\n", + " -0.068094\n", " \n", " \n", " 4\n", @@ -2598,16 +3052,16 @@ " 163840\n", " 170729\n", " ...\n", - " -0.002933\n", - " -0.004738\n", - " -0.003541\n", - " -0.001943\n", - " -0.002958\n", - " -0.001370\n", - " -0.002272\n", - " -0.003326\n", - " -0.002195\n", - " -0.003648\n", + " 0.152879\n", + " 0.291377\n", + " -0.198813\n", + " -0.429966\n", + " 0.270807\n", + " -0.239578\n", + " 0.068387\n", + " -0.406068\n", + " -0.293937\n", + " -0.101793\n", " \n", " \n", " ...\n", @@ -2646,16 +3100,16 @@ " 163840\n", " 227488\n", " ...\n", - " -0.004391\n", - " 0.004506\n", - " -0.017141\n", - " 0.000025\n", - " 0.042810\n", - " 0.024843\n", - " -0.022073\n", - " 0.043973\n", - " 0.005934\n", - " -0.006544\n", + " -0.391025\n", + " -0.056139\n", + " -0.131877\n", + " 0.002086\n", + " 0.052237\n", + " -0.073272\n", + " -0.171256\n", + " 0.151294\n", + " -0.122621\n", + " 0.059819\n", " \n", " \n", " 78\n", @@ -2670,16 +3124,16 @@ " 163840\n", " 227488\n", " ...\n", - " -0.129386\n", - " -0.144098\n", - " -0.077484\n", - " -0.091391\n", - " -0.104009\n", - " -0.034113\n", - " -0.096020\n", - " -0.063028\n", - " -0.079460\n", - " -0.102111\n", + " -0.026493\n", + " -0.118427\n", + " 0.084949\n", + " -0.022378\n", + " -0.210016\n", + " -0.091448\n", + " 0.038883\n", + " -0.119953\n", + " -0.181654\n", + " -0.168288\n", " \n", " \n", " 79\n", @@ -2694,16 +3148,16 @@ " 163840\n", " 174642\n", " ...\n", - " 0.000218\n", - " 0.000532\n", - " 0.000263\n", - " 0.000122\n", - " 0.000352\n", - " 0.000184\n", - " 0.000139\n", - " 0.000887\n", - " 0.000173\n", - " 0.000558\n", + " -0.414233\n", + " -0.028989\n", + " -0.079399\n", + " 0.069994\n", + " 0.022942\n", + " -0.047608\n", + " -0.206531\n", + " 0.133008\n", + " -0.047079\n", + " -0.011383\n", " \n", " \n", " 80\n", @@ -2718,16 +3172,16 @@ " 163840\n", " 174642\n", " ...\n", - " -0.000218\n", - " -0.000109\n", - " -0.000204\n", - " 0.000009\n", - " -0.000265\n", - " -0.000104\n", - " -0.000189\n", - " 0.000523\n", - " -0.000149\n", - " -0.000385\n", + " -0.084961\n", + " -0.115746\n", + " -0.085654\n", + " -0.038717\n", + " 0.006230\n", + " -0.039447\n", + " -0.157264\n", + " -0.002997\n", + " -0.075576\n", + " -0.172018\n", " \n", " \n", " 81\n", @@ -2742,16 +3196,16 @@ " 163840\n", " 174642\n", " ...\n", - " -0.001127\n", - " -0.002115\n", - " -0.001278\n", - " -0.000581\n", - " -0.001344\n", - " -0.000724\n", - " -0.000569\n", - " -0.003553\n", - " -0.000857\n", - " -0.002281\n", + " -0.223092\n", + " -0.099597\n", + " -0.119387\n", + " 0.070661\n", + " -0.016427\n", + " -0.110365\n", + " -0.208193\n", + " 0.088656\n", + " -0.018158\n", + " 0.014191\n", " \n", " \n", "\n", @@ -2773,35 +3227,35 @@ "81 chr1 109728807 T GG ALX3 109710224 110234512 - \n", "\n", " gene_mask_start gene_mask_end ... agg_9528 agg_9529 agg_9530 \\\n", - "0 163840 172672 ... -0.003487 -0.006149 -0.003175 \n", - "1 163840 172672 ... -0.000096 -0.000190 -0.000135 \n", - "2 163840 178946 ... -0.002595 -0.004256 -0.002759 \n", - "3 163840 178946 ... -0.000587 -0.000894 -0.000563 \n", - "4 163840 170729 ... -0.002933 -0.004738 -0.003541 \n", + "0 163840 172672 ... -0.006802 -0.043372 -0.031292 \n", + "1 163840 172672 ... -0.245593 -0.190111 -0.137129 \n", + "2 163840 178946 ... -0.005441 -0.041043 -0.028593 \n", + "3 163840 178946 ... -0.241628 -0.187019 -0.134874 \n", + "4 163840 170729 ... 0.152879 0.291377 -0.198813 \n", ".. ... ... ... ... ... ... \n", - "77 163840 227488 ... -0.004391 0.004506 -0.017141 \n", - "78 163840 227488 ... -0.129386 -0.144098 -0.077484 \n", - "79 163840 174642 ... 0.000218 0.000532 0.000263 \n", - "80 163840 174642 ... -0.000218 -0.000109 -0.000204 \n", - "81 163840 174642 ... -0.001127 -0.002115 -0.001278 \n", + "77 163840 227488 ... -0.391025 -0.056139 -0.131877 \n", + "78 163840 227488 ... -0.026493 -0.118427 0.084949 \n", + "79 163840 174642 ... -0.414233 -0.028989 -0.079399 \n", + "80 163840 174642 ... -0.084961 -0.115746 -0.085654 \n", + "81 163840 174642 ... -0.223092 -0.099597 -0.119387 \n", "\n", " agg_9531 agg_9532 agg_9533 agg_9535 agg_9536 agg_9537 agg_9538 \n", - "0 -0.002726 -0.003147 -0.001746 -0.002291 -0.005283 -0.003393 -0.006694 \n", - "1 -0.000043 -0.000076 -0.000072 -0.000106 0.000029 -0.000055 -0.000119 \n", - "2 -0.001601 -0.002601 -0.001238 -0.001756 -0.002918 -0.001914 -0.003244 \n", - "3 -0.000324 -0.000471 -0.000258 -0.000382 -0.000683 -0.000459 -0.000757 \n", - "4 -0.001943 -0.002958 -0.001370 -0.002272 -0.003326 -0.002195 -0.003648 \n", + "0 -0.002179 0.046106 -0.025365 -0.086671 0.013381 -0.037975 -0.062525 \n", + "1 0.051578 0.003079 -0.128517 -0.236761 0.017629 -0.094391 -0.070108 \n", + "2 0.001764 0.048425 -0.024294 -0.086048 0.018201 -0.035142 -0.059727 \n", + "3 0.050618 0.005003 -0.125120 -0.233175 0.019390 -0.093390 -0.068094 \n", + "4 -0.429966 0.270807 -0.239578 0.068387 -0.406068 -0.293937 -0.101793 \n", ".. ... ... ... ... ... ... ... \n", - "77 0.000025 0.042810 0.024843 -0.022073 0.043973 0.005934 -0.006544 \n", - "78 -0.091391 -0.104009 -0.034113 -0.096020 -0.063028 -0.079460 -0.102111 \n", - "79 0.000122 0.000352 0.000184 0.000139 0.000887 0.000173 0.000558 \n", - "80 0.000009 -0.000265 -0.000104 -0.000189 0.000523 -0.000149 -0.000385 \n", - "81 -0.000581 -0.001344 -0.000724 -0.000569 -0.003553 -0.000857 -0.002281 \n", + "77 0.002086 0.052237 -0.073272 -0.171256 0.151294 -0.122621 0.059819 \n", + "78 -0.022378 -0.210016 -0.091448 0.038883 -0.119953 -0.181654 -0.168288 \n", + "79 0.069994 0.022942 -0.047608 -0.206531 0.133008 -0.047079 -0.011383 \n", + "80 -0.038717 0.006230 -0.039447 -0.157264 -0.002997 -0.075576 -0.172018 \n", + "81 0.070661 -0.016427 -0.110365 -0.208193 0.088656 -0.018158 0.014191 \n", "\n", "[82 rows x 8870 columns]" ] }, - "execution_count": 15, + "execution_count": 18, "metadata": {}, "output_type": "execute_result" } @@ -2814,47 +3268,58 @@ "cell_type": "markdown", "metadata": {}, "source": [ - "Or variant effect can be performed on vcf file." + "Or variant effect prediction can be performed on a vcf file." ] }, { "cell_type": "code", - "execution_count": 16, + "execution_count": 19, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact decima_metadata:latest, 628.05MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:0.5 (1272.2MB/s)\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact decima_rep0:latest, 2155.88MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:1.2 (1862.4MB/s)\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact human_state_dict_fold0:latest, 709.30MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:0.5 (1514.0MB/s)\n", - "Using default `ModelCheckpoint`. Consider installing `litmodels` package to enable `LitModelCheckpoint` for automatic upload to the Lightning model registry.\n", + "/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/lightning_fabric/plugins/environments/slurm.py:204: PossibleUserWarning: The `srun` command is available on your system but is not used. HINT: If your intention is to run Lightning on SLURM, prepend your python command with `srun` like so: srun python /gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib ...\n", "GPU available: True (cuda), used: True\n", "TPU available: False, using: 0 TPU cores\n", - "HPU available: False, using: 0 HPUs\n", + "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n", + "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n", "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n" ] }, { "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "6c19dcc02b5f43c89ea51643583f352d", - "version_major": 2, - "version_minor": 0 - }, + "text/html": [ + "
/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/rich/live.py:260: UserWarning: \n",
+       "install \"ipywidgets\" for Jupyter support\n",
+       "
\n" + ], "text/plain": [ - "Predicting: | | 0/? [00:00\n" + ], + "text/plain": [] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "name": "stderr", + "output_type": "stream", + "text": [ + "Warnings:\n", + "allele_mismatch_with_reference_genome: 10 alleles out of 48 predictions mismatched with the genome file /home/lala8/.local/share/genomes/hg38/hg38.fa. If this is not expected, please check if you are using the correct genome version.\n" + ] } ], "source": [ @@ -2863,7 +3328,7 @@ }, { "cell_type": "code", - "execution_count": 17, + "execution_count": 20, "metadata": {}, "outputs": [ { @@ -2924,16 +3389,16 @@ " 163840\n", " 172672\n", " ...\n", - " -0.000096\n", - " -0.000190\n", - " -0.000135\n", - " -0.000043\n", - " -0.000076\n", - " -0.000072\n", - " -0.000106\n", - " 0.000029\n", - " -0.000055\n", - " -0.000119\n", + " -0.038816\n", + " -0.132628\n", + " 0.012945\n", + " -0.057034\n", + " -0.045095\n", + " -0.019845\n", + " -0.048861\n", + " -0.248287\n", + " -0.157361\n", + " -0.202191\n", " \n", " \n", " 1\n", @@ -2948,16 +3413,16 @@ " 163840\n", " 178946\n", " ...\n", - " -0.000587\n", - " -0.000894\n", - " -0.000563\n", - " -0.000324\n", - " -0.000471\n", - " -0.000258\n", - " -0.000382\n", - " -0.000683\n", - " -0.000459\n", - " -0.000757\n", + " 0.007422\n", + " -0.040294\n", + " 0.037184\n", + " -0.039102\n", + " -0.069273\n", + " -0.089758\n", + " 0.085668\n", + " -0.100051\n", + " 0.199948\n", + " -0.228135\n", " \n", " \n", " 2\n", @@ -2972,16 +3437,16 @@ " 163840\n", " 170729\n", " ...\n", - " -0.000845\n", - " -0.001181\n", - " -0.000784\n", - " -0.000442\n", - " -0.000589\n", - " -0.000313\n", - " -0.000526\n", - " -0.000856\n", - " -0.000637\n", - " -0.000881\n", + " -0.038910\n", + " -0.132768\n", + " 0.012589\n", + " -0.057297\n", + " -0.044415\n", + " -0.019266\n", + " -0.049203\n", + " -0.247513\n", + " -0.158220\n", + " -0.202740\n", " \n", " \n", " 3\n", @@ -2996,16 +3461,16 @@ " 163840\n", " 165050\n", " ...\n", - " -0.001767\n", - " -0.002189\n", - " -0.001705\n", - " -0.001091\n", - " -0.001184\n", - " -0.000717\n", - " -0.001306\n", - " -0.001550\n", - " -0.001349\n", - " -0.001550\n", + " 0.006907\n", + " -0.041188\n", + " 0.036539\n", + " -0.038984\n", + " -0.070001\n", + " -0.090065\n", + " 0.085498\n", + " -0.100311\n", + " 0.199196\n", + " -0.229159\n", " \n", " \n", " 4\n", @@ -3020,16 +3485,16 @@ " 163840\n", " 166306\n", " ...\n", - " -0.000098\n", - " 0.000052\n", - " -0.000040\n", - " -0.000051\n", - " 0.000037\n", - " 0.000028\n", - " -0.000024\n", - " 0.000093\n", - " -0.000080\n", - " -0.000203\n", + " 0.384163\n", + " 0.383833\n", + " 0.208923\n", + " 0.197689\n", + " 0.147001\n", + " 0.214184\n", + " 0.429241\n", + " 0.046892\n", + " 0.232819\n", + " 0.122424\n", " \n", " \n", " 5\n", @@ -3044,16 +3509,16 @@ " 163840\n", " 167179\n", " ...\n", - " -0.000603\n", - " -0.000860\n", - " -0.000584\n", - " -0.000328\n", - " -0.000361\n", - " -0.000211\n", - " -0.000395\n", - " -0.000428\n", - " -0.000371\n", - " -0.000468\n", + " 0.128392\n", + " 0.201155\n", + " 0.096084\n", + " 0.117298\n", + " -0.126285\n", + " -0.059694\n", + " 0.184677\n", + " -0.009753\n", + " 0.305165\n", + " -0.107976\n", " \n", " \n", " 6\n", @@ -3068,16 +3533,16 @@ " 163840\n", " 198383\n", " ...\n", - " -0.000358\n", - " -0.000528\n", - " -0.000338\n", - " -0.000193\n", - " -0.000205\n", - " -0.000113\n", - " -0.000245\n", - " -0.000265\n", - " -0.000241\n", - " -0.000347\n", + " 0.382839\n", + " 0.387868\n", + " 0.207169\n", + " 0.192711\n", + " 0.148570\n", + " 0.212126\n", + " 0.429055\n", + " 0.040387\n", + " 0.231833\n", + " 0.117729\n", " \n", " \n", " 7\n", @@ -3092,16 +3557,16 @@ " 163840\n", " 184493\n", " ...\n", - " 0.001338\n", - " 0.000849\n", - " 0.000867\n", - " 0.000903\n", - " 0.001198\n", - " 0.001515\n", - " 0.001150\n", - " 0.001019\n", - " 0.000498\n", - " 0.001034\n", + " 0.127002\n", + " 0.199668\n", + " 0.092870\n", + " 0.113858\n", + " -0.132513\n", + " -0.059923\n", + " 0.180629\n", + " -0.008512\n", + " 0.302857\n", + " -0.112850\n", " \n", " \n", " 8\n", @@ -3116,16 +3581,16 @@ " 163840\n", " 168975\n", " ...\n", - " -0.000518\n", - " -0.000364\n", - " -0.000522\n", - " -0.000307\n", - " -0.000214\n", - " -0.000158\n", - " -0.000330\n", - " -0.000363\n", - " -0.000348\n", - " -0.000509\n", + " -0.065035\n", + " -0.009808\n", + " -0.089304\n", + " -0.123990\n", + " -0.237687\n", + " -0.145043\n", + " -0.027265\n", + " -0.172405\n", + " -0.209000\n", + " -0.169588\n", " \n", " \n", " 9\n", @@ -3140,16 +3605,16 @@ " 163840\n", " 173223\n", " ...\n", - " 0.000109\n", - " 0.000011\n", - " 0.000054\n", - " 0.000009\n", - " -0.000056\n", - " -0.000199\n", - " -0.000207\n", - " -0.000112\n", - " 0.000043\n", - " 0.000051\n", + " 0.149715\n", + " 0.036652\n", + " 0.236477\n", + " -0.033082\n", + " -0.059123\n", + " -0.109544\n", + " -0.116550\n", + " -0.056748\n", + " 0.238428\n", + " -0.066090\n", " \n", " \n", " 10\n", @@ -3164,16 +3629,16 @@ " 163840\n", " 170339\n", " ...\n", - " -0.000278\n", - " -0.000460\n", - " -0.000305\n", - " -0.000142\n", - " -0.000220\n", - " -0.000106\n", - " -0.000170\n", - " -0.000272\n", - " -0.000170\n", - " -0.000231\n", + " -0.064135\n", + " -0.008839\n", + " -0.088697\n", + " -0.123552\n", + " -0.236969\n", + " -0.144438\n", + " -0.026115\n", + " -0.171792\n", + " -0.208494\n", + " -0.170012\n", " \n", " \n", " 11\n", @@ -3188,16 +3653,16 @@ " 163840\n", " 177242\n", " ...\n", - " 0.007892\n", - " 0.005266\n", - " 0.008324\n", - " 0.007764\n", - " -0.001580\n", - " 0.002468\n", - " 0.008656\n", - " 0.002925\n", - " 0.007485\n", - " 0.010022\n", + " 0.149233\n", + " 0.036425\n", + " 0.236163\n", + " -0.033192\n", + " -0.059114\n", + " -0.109789\n", + " -0.116972\n", + " -0.057177\n", + " 0.238166\n", + " -0.067322\n", " \n", " \n", " 12\n", @@ -3212,16 +3677,16 @@ " 163840\n", " 166924\n", " ...\n", - " -0.000308\n", - " -0.000433\n", - " -0.000336\n", - " -0.000197\n", - " -0.000195\n", - " -0.000144\n", - " -0.000220\n", - " -0.000190\n", - " -0.000199\n", - " -0.000302\n", + " 0.025078\n", + " 0.316090\n", + " 0.052550\n", + " -0.128108\n", + " -0.025624\n", + " 0.018586\n", + " 0.070953\n", + " 0.035494\n", + " 0.035036\n", + " 0.198598\n", " \n", " \n", " 13\n", @@ -3236,16 +3701,16 @@ " 163840\n", " 166653\n", " ...\n", - " -0.000255\n", - " -0.000385\n", - " -0.000261\n", - " -0.000158\n", - " -0.000164\n", - " -0.000123\n", - " -0.000158\n", - " -0.000281\n", - " -0.000191\n", - " -0.000319\n", + " -0.060781\n", + " 0.136466\n", + " 0.222455\n", + " 0.180580\n", + " -0.077465\n", + " 0.059501\n", + " -0.152240\n", + " 0.021208\n", + " 0.388104\n", + " 0.088033\n", " \n", " \n", " 14\n", @@ -3260,16 +3725,16 @@ " 163840\n", " 199838\n", " ...\n", - " -0.000901\n", - " -0.001054\n", - " -0.000771\n", - " -0.001977\n", - " -0.000437\n", - " -0.000555\n", - " -0.001084\n", - " -0.000583\n", - " -0.001613\n", - " -0.000536\n", + " 0.024326\n", + " 0.315336\n", + " 0.050626\n", + " -0.121845\n", + " -0.024810\n", + " 0.020802\n", + " 0.065568\n", + " 0.039766\n", + " 0.038300\n", + " 0.207067\n", " \n", " \n", " 15\n", @@ -3284,16 +3749,16 @@ " 163840\n", " 178999\n", " ...\n", - " -0.000146\n", - " -0.000230\n", - " -0.000180\n", - " -0.000099\n", - " -0.000132\n", - " -0.000082\n", - " -0.000092\n", - " -0.000189\n", - " -0.000112\n", - " -0.000195\n", + " -0.069126\n", + " 0.131456\n", + " 0.215256\n", + " 0.164175\n", + " -0.088568\n", + " 0.049765\n", + " -0.160795\n", + " 0.012626\n", + " 0.377364\n", + " 0.080307\n", " \n", " \n", " 16\n", @@ -3308,16 +3773,16 @@ " 163840\n", " 168116\n", " ...\n", - " -0.000458\n", - " -0.000697\n", - " -0.000491\n", - " -0.000359\n", - " -0.000367\n", - " -0.000241\n", - " -0.000268\n", - " -0.000487\n", - " -0.000344\n", - " -0.000595\n", + " -0.068721\n", + " -0.131200\n", + " -0.095311\n", + " 0.046062\n", + " 0.058227\n", + " -0.014569\n", + " -0.110806\n", + " 0.156508\n", + " 0.050760\n", + " 0.086102\n", " \n", " \n", " 17\n", @@ -3332,16 +3797,16 @@ " 163840\n", " 183816\n", " ...\n", - " -0.000815\n", - " -0.001141\n", - " -0.000897\n", - " -0.000674\n", - " -0.000717\n", - " -0.000515\n", - " -0.000520\n", - " -0.000780\n", - " -0.000614\n", - " -0.001010\n", + " -0.000722\n", + " -0.065477\n", + " -0.015448\n", + " -0.048512\n", + " 0.080478\n", + " 0.044108\n", + " -0.040190\n", + " -0.109651\n", + " 0.016822\n", + " -0.062620\n", " \n", " \n", " 18\n", @@ -3356,16 +3821,16 @@ " 163840\n", " 181059\n", " ...\n", - " -0.000748\n", - " -0.001163\n", - " -0.000843\n", - " -0.000615\n", - " -0.000704\n", - " -0.000456\n", - " -0.000404\n", - " -0.000761\n", - " -0.000572\n", - " -0.000986\n", + " -0.068234\n", + " -0.130586\n", + " -0.095205\n", + " 0.046179\n", + " 0.059056\n", + " -0.013978\n", + " -0.110189\n", + " 0.157281\n", + " 0.050832\n", + " 0.085974\n", " \n", " \n", " 19\n", @@ -3380,16 +3845,16 @@ " 163840\n", " 176946\n", " ...\n", - " 0.000021\n", - " 0.000008\n", - " 0.000037\n", - " 0.000035\n", - " 0.000096\n", - " 0.000057\n", - " 0.000009\n", - " 0.000184\n", - " 0.000039\n", - " 0.000044\n", + " -0.000299\n", + " -0.064919\n", + " -0.015113\n", + " -0.048239\n", + " 0.080818\n", + " 0.044669\n", + " -0.040059\n", + " -0.109231\n", + " 0.017035\n", + " -0.062525\n", " \n", " \n", " 20\n", @@ -3404,16 +3869,16 @@ " 163840\n", " 177982\n", " ...\n", - " -0.000325\n", - " -0.000491\n", - " -0.000359\n", - " -0.000260\n", - " -0.000322\n", - " -0.000207\n", - " -0.000179\n", - " -0.000336\n", - " -0.000235\n", - " -0.000382\n", + " 0.144722\n", + " 0.096918\n", + " 0.053467\n", + " 0.178367\n", + " 0.186895\n", + " -0.086145\n", + " 0.078297\n", + " 0.339905\n", + " 0.164013\n", + " 0.351803\n", " \n", " \n", " 21\n", @@ -3428,16 +3893,16 @@ " 163840\n", " 172928\n", " ...\n", - " -0.000077\n", - " -0.000098\n", - " -0.000090\n", - " -0.000067\n", - " -0.000092\n", - " -0.000057\n", - " -0.000034\n", - " -0.000212\n", - " -0.000063\n", - " -0.000159\n", + " -0.204408\n", + " -0.164630\n", + " -0.161219\n", + " -0.077524\n", + " 0.001738\n", + " -0.079267\n", + " -0.283000\n", + " -0.061010\n", + " -0.083843\n", + " -0.061747\n", " \n", " \n", " 22\n", @@ -3452,16 +3917,16 @@ " 163840\n", " 180515\n", " ...\n", - " 0.000103\n", - " 0.000070\n", - " 0.000060\n", - " 0.000038\n", - " 0.000027\n", - " 0.000055\n", - " 0.000078\n", - " 0.000277\n", - " 0.000062\n", - " 0.000245\n", + " 0.145138\n", + " 0.099315\n", + " 0.055320\n", + " 0.177228\n", + " 0.190383\n", + " -0.084195\n", + " 0.083160\n", + " 0.344412\n", + " 0.166400\n", + " 0.354686\n", " \n", " \n", " 23\n", @@ -3476,16 +3941,16 @@ " 163840\n", " 180515\n", " ...\n", - " 0.003317\n", - " 0.005164\n", - " 0.002698\n", - " 0.001438\n", - " 0.001558\n", - " 0.001273\n", - " 0.001937\n", - " 0.005484\n", - " 0.002190\n", - " 0.005321\n", + " -0.216383\n", + " -0.174637\n", + " -0.171691\n", + " -0.086247\n", + " -0.017797\n", + " -0.098000\n", + " -0.294861\n", + " -0.081399\n", + " -0.096354\n", + " -0.078279\n", " \n", " \n", " 24\n", @@ -3500,16 +3965,16 @@ " 163840\n", " 179428\n", " ...\n", - " 0.002225\n", - " 0.001688\n", - " 0.001366\n", - " 0.001251\n", - " 0.001342\n", - " 0.000177\n", - " 0.001402\n", - " 0.000480\n", - " 0.001663\n", - " 0.001764\n", + " 0.125082\n", + " 0.187445\n", + " 0.094098\n", + " 0.073541\n", + " 0.153524\n", + " 0.122184\n", + " 0.178186\n", + " 0.325020\n", + " 0.131711\n", + " 0.251710\n", " \n", " \n", " 25\n", @@ -3524,16 +3989,16 @@ " 163840\n", " 179428\n", " ...\n", - " -0.004184\n", - " -0.003700\n", - " -0.003180\n", - " -0.004169\n", - " -0.003774\n", - " -0.003399\n", - " -0.003382\n", - " 0.000949\n", - " -0.002237\n", - " -0.002981\n", + " -0.072443\n", + " -0.081550\n", + " -0.038486\n", + " -0.047714\n", + " 0.226176\n", + " 0.045261\n", + " -0.126760\n", + " 0.371777\n", + " 0.060598\n", + " -0.005164\n", " \n", " \n", " 26\n", @@ -3548,16 +4013,16 @@ " 163840\n", " 173165\n", " ...\n", - " -0.000055\n", - " -0.000036\n", - " 0.000088\n", - " -0.000085\n", - " 0.000037\n", - " -0.000040\n", - " -0.000056\n", - " -0.000044\n", - " -0.000046\n", - " -0.000027\n", + " 0.125406\n", + " 0.187519\n", + " 0.094387\n", + " 0.074029\n", + " 0.154567\n", + " 0.122893\n", + " 0.178522\n", + " 0.326261\n", + " 0.131973\n", + " 0.251846\n", " \n", " \n", " 27\n", @@ -3572,16 +4037,16 @@ " 163840\n", " 173165\n", " ...\n", - " 0.000644\n", - " -0.000063\n", - " 0.001706\n", - " 0.001486\n", - " 0.003162\n", - " 0.001671\n", - " 0.000299\n", - " 0.002102\n", - " 0.001007\n", - " 0.001831\n", + " -0.072282\n", + " -0.081517\n", + " -0.038386\n", + " -0.047584\n", + " 0.226377\n", + " 0.045594\n", + " -0.126571\n", + " 0.372053\n", + " 0.060832\n", + " -0.004993\n", " \n", " \n", " 28\n", @@ -3596,16 +4061,16 @@ " 163840\n", " 172720\n", " ...\n", - " 0.000533\n", - " 0.000539\n", - " 0.000585\n", - " 0.000474\n", - " 0.000076\n", - " 0.000280\n", - " 0.000422\n", - " -0.000033\n", - " 0.000436\n", - " 0.000369\n", + " -0.109378\n", + " -0.095939\n", + " -0.046820\n", + " -0.160233\n", + " -0.108172\n", + " -0.116325\n", + " 0.049603\n", + " 0.171065\n", + " 0.111874\n", + " 0.250584\n", " \n", " \n", " 29\n", @@ -3620,16 +4085,16 @@ " 163840\n", " 172720\n", " ...\n", - " 0.003902\n", - " 0.002963\n", - " 0.001609\n", - " 0.002841\n", - " 0.000024\n", - " 0.000726\n", - " 0.001582\n", - " 0.000497\n", - " 0.002237\n", - " -0.000092\n", + " -0.014295\n", + " -0.047262\n", + " 0.040566\n", + " 0.034287\n", + " 0.086552\n", + " -0.002746\n", + " -0.036511\n", + " 0.105897\n", + " 0.075976\n", + " -0.094442\n", " \n", " \n", " 30\n", @@ -3644,16 +4109,16 @@ " 163840\n", " 172374\n", " ...\n", - " 0.000144\n", - " 0.000244\n", - " 0.000101\n", - " 0.000130\n", - " -0.000119\n", - " -0.000066\n", - " 0.000112\n", - " 0.000059\n", - " 0.000113\n", - " 0.000135\n", + " -0.114771\n", + " -0.101728\n", + " -0.051283\n", + " -0.167664\n", + " -0.115162\n", + " -0.118221\n", + " 0.045587\n", + " 0.164287\n", + " 0.105892\n", + " 0.242462\n", " \n", " \n", " 31\n", @@ -3668,16 +4133,16 @@ " 163840\n", " 172374\n", " ...\n", - " 0.000287\n", - " 0.000869\n", - " 0.000338\n", - " 0.000516\n", - " 0.000702\n", - " 0.000260\n", - " 0.000547\n", - " 0.000815\n", - " 0.000190\n", - " 0.000561\n", + " -0.014773\n", + " -0.051328\n", + " 0.039865\n", + " 0.034642\n", + " 0.079301\n", + " -0.005801\n", + " -0.037709\n", + " 0.102587\n", + " 0.075023\n", + " -0.093098\n", " \n", " \n", " 32\n", @@ -3692,16 +4157,16 @@ " 163840\n", " 170946\n", " ...\n", - " -0.000685\n", - " -0.000671\n", - " -0.000490\n", - " -0.000305\n", - " -0.000203\n", - " -0.000045\n", - " -0.000183\n", - " -0.000423\n", - " -0.000412\n", - " -0.000544\n", + " 0.017597\n", + " -0.080504\n", + " 0.049995\n", + " -0.011808\n", + " -0.271627\n", + " -0.061634\n", + " -0.068092\n", + " -0.169892\n", + " -0.139209\n", + " -0.134266\n", " \n", " \n", " 33\n", @@ -3716,16 +4181,16 @@ " 163840\n", " 170946\n", " ...\n", - " 0.014859\n", - " 0.029183\n", - " 0.015364\n", - " 0.007996\n", - " 0.015255\n", - " 0.007950\n", - " 0.010278\n", - " 0.021851\n", - " 0.009230\n", - " 0.024402\n", + " -0.297318\n", + " 0.036640\n", + " 0.039409\n", + " 0.172793\n", + " 0.131960\n", + " 0.099102\n", + " -0.127269\n", + " 0.248293\n", + " 0.078142\n", + " 0.244484\n", " \n", " \n", " 34\n", @@ -3740,16 +4205,16 @@ " 163840\n", " 233549\n", " ...\n", - " 0.002294\n", - " 0.002954\n", - " 0.003762\n", - " 0.001515\n", - " -0.001879\n", - " -0.003638\n", - " 0.004167\n", - " -0.000977\n", - " 0.003568\n", - " 0.001709\n", + " 0.017213\n", + " -0.081016\n", + " 0.048742\n", + " -0.012799\n", + " -0.271882\n", + " -0.061404\n", + " -0.068229\n", + " -0.168541\n", + " -0.140811\n", + " -0.135465\n", " \n", " \n", " 35\n", @@ -3764,16 +4229,16 @@ " 163840\n", " 233549\n", " ...\n", - " -0.002409\n", - " -0.002181\n", - " 0.001515\n", - " 0.007504\n", - " 0.011276\n", - " 0.019385\n", - " -0.002945\n", - " 0.020405\n", - " -0.001569\n", - " 0.007003\n", + " -0.296310\n", + " 0.027849\n", + " 0.034702\n", + " 0.172035\n", + " 0.133321\n", + " 0.101695\n", + " -0.133212\n", + " 0.247434\n", + " 0.074210\n", + " 0.243297\n", " \n", " \n", " 36\n", @@ -3788,16 +4253,16 @@ " 163840\n", " 179789\n", " ...\n", - " 0.000565\n", - " 0.001265\n", - " 0.001679\n", - " -0.000034\n", - " -0.002686\n", - " -0.000878\n", - " 0.001329\n", - " -0.001112\n", - " 0.002059\n", - " 0.000288\n", + " 0.173036\n", + " 0.238832\n", + " -0.112437\n", + " -0.396939\n", + " -0.048346\n", + " -0.349238\n", + " 0.178761\n", + " -0.541497\n", + " -0.432859\n", + " -0.115770\n", " \n", " \n", " 37\n", @@ -3812,16 +4277,16 @@ " 163840\n", " 179789\n", " ...\n", - " 0.015521\n", - " 0.005221\n", - " 0.011660\n", - " 0.028807\n", - " 0.010368\n", - " 0.014050\n", - " 0.007810\n", - " 0.026553\n", - " 0.009914\n", - " 0.026500\n", + " -0.587394\n", + " -0.383097\n", + " -0.074972\n", + " 0.116038\n", + " -0.106850\n", + " -0.536714\n", + " -0.289871\n", + " 0.275144\n", + " -0.386951\n", + " -0.475129\n", " \n", " \n", " 38\n", @@ -3836,16 +4301,16 @@ " 163840\n", " 182577\n", " ...\n", - " -0.000140\n", - " 0.001370\n", - " 0.000755\n", - " 0.000528\n", - " -0.000446\n", - " -0.000184\n", - " -0.001062\n", - " 0.000593\n", - " 0.000575\n", - " 0.002940\n", + " 0.196702\n", + " 0.244239\n", + " -0.096241\n", + " -0.383522\n", + " -0.053860\n", + " -0.346986\n", + " 0.204916\n", + " -0.547306\n", + " -0.428087\n", + " -0.109396\n", " \n", " \n", " 39\n", @@ -3860,16 +4325,16 @@ " 163840\n", " 182577\n", " ...\n", - " -0.032480\n", - " -0.033616\n", - " -0.029502\n", - " -0.050562\n", - " -0.030190\n", - " -0.021981\n", - " -0.012192\n", - " -0.029633\n", - " -0.037296\n", - " -0.067220\n", + " -0.598326\n", + " -0.350383\n", + " -0.062013\n", + " 0.137351\n", + " -0.113740\n", + " -0.549226\n", + " -0.316263\n", + " 0.282046\n", + " -0.367199\n", + " -0.457674\n", " \n", " \n", " 40\n", @@ -3884,16 +4349,16 @@ " 163840\n", " 205369\n", " ...\n", - " -0.001281\n", - " -0.001583\n", - " -0.000117\n", - " -0.000501\n", - " -0.001837\n", - " -0.001011\n", - " -0.002222\n", - " 0.000031\n", - " 0.000068\n", - " 0.000062\n", + " -0.065015\n", + " -0.134536\n", + " -0.140023\n", + " 0.024319\n", + " -0.060507\n", + " -0.108153\n", + " -0.149791\n", + " -0.038775\n", + " -0.076944\n", + " -0.118178\n", " \n", " \n", " 41\n", @@ -3908,16 +4373,16 @@ " 163840\n", " 205369\n", " ...\n", - " 0.006389\n", - " 0.010391\n", - " 0.008292\n", - " 0.010647\n", - " -0.005511\n", - " 0.002638\n", - " -0.004853\n", - " -0.009829\n", - " 0.008542\n", - " 0.011590\n", + " -0.138112\n", + " -0.135524\n", + " -0.122887\n", + " 0.003460\n", + " -0.095921\n", + " -0.142851\n", + " -0.133408\n", + " -0.044298\n", + " -0.079864\n", + " 0.063350\n", " \n", " \n", " 42\n", @@ -3932,16 +4397,16 @@ " 163840\n", " 185065\n", " ...\n", - " 0.001077\n", - " 0.002305\n", - " 0.001138\n", - " -0.000056\n", - " -0.000738\n", - " 0.000059\n", - " 0.000456\n", - " -0.000396\n", - " 0.000904\n", - " 0.000512\n", + " -0.064936\n", + " -0.134381\n", + " -0.140410\n", + " 0.024382\n", + " -0.060211\n", + " -0.106950\n", + " -0.150407\n", + " -0.038308\n", + " -0.077518\n", + " -0.118339\n", " \n", " \n", " 43\n", @@ -3956,16 +4421,16 @@ " 163840\n", " 185065\n", " ...\n", - " 0.009587\n", - " 0.012155\n", - " 0.002769\n", - " 0.018908\n", - " 0.010602\n", - " 0.000302\n", - " 0.003231\n", - " 0.007630\n", - " 0.011715\n", - " 0.019595\n", + " -0.138833\n", + " -0.136745\n", + " -0.123534\n", + " 0.002528\n", + " -0.096342\n", + " -0.143056\n", + " -0.133687\n", + " -0.044767\n", + " -0.080880\n", + " 0.062666\n", " \n", " \n", " 44\n", @@ -3980,16 +4445,16 @@ " 163840\n", " 227488\n", " ...\n", - " 0.001978\n", - " 0.001395\n", - " 0.001501\n", - " 0.000012\n", - " -0.000538\n", - " -0.000914\n", - " 0.002817\n", - " -0.001998\n", - " 0.001149\n", - " 0.002069\n", + " -0.260044\n", + " -0.168338\n", + " -0.271208\n", + " -0.247191\n", + " -0.235198\n", + " -0.383056\n", + " -0.163678\n", + " -0.055487\n", + " -0.088683\n", + " -0.155701\n", " \n", " \n", " 45\n", @@ -4004,16 +4469,16 @@ " 163840\n", " 227488\n", " ...\n", - " -0.004803\n", - " 0.004060\n", - " -0.018828\n", - " -0.000383\n", - " 0.042374\n", - " 0.022364\n", - " -0.023450\n", - " 0.043353\n", - " 0.006033\n", - " -0.006614\n", + " 0.107979\n", + " 0.020519\n", + " 0.216325\n", + " 0.572342\n", + " 0.147766\n", + " 0.106553\n", + " 0.096116\n", + " 0.279162\n", + " 0.162525\n", + " 0.332815\n", " \n", " \n", " 46\n", @@ -4028,16 +4493,16 @@ " 163840\n", " 174642\n", " ...\n", - " 0.000218\n", - " 0.000532\n", - " 0.000263\n", - " 0.000122\n", - " 0.000352\n", - " 0.000184\n", - " 0.000139\n", - " 0.000887\n", - " 0.000173\n", - " 0.000558\n", + " -0.256797\n", + " -0.163505\n", + " -0.266977\n", + " -0.247593\n", + " -0.233061\n", + " -0.386286\n", + " -0.155122\n", + " -0.059363\n", + " -0.081943\n", + " -0.156882\n", " \n", " \n", " 47\n", @@ -4052,16 +4517,16 @@ " 163840\n", " 174642\n", " ...\n", - " -0.000386\n", - " -0.000480\n", - " -0.000415\n", - " -0.000092\n", - " -0.000513\n", - " -0.000241\n", - " -0.000285\n", - " -0.000141\n", - " -0.000277\n", - " -0.000758\n", + " 0.100775\n", + " 0.018718\n", + " 0.213268\n", + " 0.574074\n", + " 0.151917\n", + " 0.110753\n", + " 0.088787\n", + " 0.286400\n", + " 0.165354\n", + " 0.329970\n", " \n", " \n", "\n", @@ -4120,109 +4585,109 @@ "47 chr1 109728807 TTT G ALX3 109710224 110234512 - \n", "\n", " gene_mask_start gene_mask_end ... agg_9528 agg_9529 agg_9530 \\\n", - "0 163840 172672 ... -0.000096 -0.000190 -0.000135 \n", - "1 163840 178946 ... -0.000587 -0.000894 -0.000563 \n", - "2 163840 170729 ... -0.000845 -0.001181 -0.000784 \n", - "3 163840 165050 ... -0.001767 -0.002189 -0.001705 \n", - "4 163840 166306 ... -0.000098 0.000052 -0.000040 \n", - "5 163840 167179 ... -0.000603 -0.000860 -0.000584 \n", - "6 163840 198383 ... -0.000358 -0.000528 -0.000338 \n", - "7 163840 184493 ... 0.001338 0.000849 0.000867 \n", - "8 163840 168975 ... -0.000518 -0.000364 -0.000522 \n", - "9 163840 173223 ... 0.000109 0.000011 0.000054 \n", - "10 163840 170339 ... -0.000278 -0.000460 -0.000305 \n", - "11 163840 177242 ... 0.007892 0.005266 0.008324 \n", - "12 163840 166924 ... -0.000308 -0.000433 -0.000336 \n", - "13 163840 166653 ... -0.000255 -0.000385 -0.000261 \n", - "14 163840 199838 ... -0.000901 -0.001054 -0.000771 \n", - "15 163840 178999 ... -0.000146 -0.000230 -0.000180 \n", - "16 163840 168116 ... -0.000458 -0.000697 -0.000491 \n", - "17 163840 183816 ... -0.000815 -0.001141 -0.000897 \n", - "18 163840 181059 ... -0.000748 -0.001163 -0.000843 \n", - "19 163840 176946 ... 0.000021 0.000008 0.000037 \n", - "20 163840 177982 ... -0.000325 -0.000491 -0.000359 \n", - "21 163840 172928 ... -0.000077 -0.000098 -0.000090 \n", - "22 163840 180515 ... 0.000103 0.000070 0.000060 \n", - "23 163840 180515 ... 0.003317 0.005164 0.002698 \n", - "24 163840 179428 ... 0.002225 0.001688 0.001366 \n", - "25 163840 179428 ... -0.004184 -0.003700 -0.003180 \n", - "26 163840 173165 ... -0.000055 -0.000036 0.000088 \n", - "27 163840 173165 ... 0.000644 -0.000063 0.001706 \n", - "28 163840 172720 ... 0.000533 0.000539 0.000585 \n", - "29 163840 172720 ... 0.003902 0.002963 0.001609 \n", - "30 163840 172374 ... 0.000144 0.000244 0.000101 \n", - "31 163840 172374 ... 0.000287 0.000869 0.000338 \n", - "32 163840 170946 ... -0.000685 -0.000671 -0.000490 \n", - "33 163840 170946 ... 0.014859 0.029183 0.015364 \n", - "34 163840 233549 ... 0.002294 0.002954 0.003762 \n", - "35 163840 233549 ... -0.002409 -0.002181 0.001515 \n", - "36 163840 179789 ... 0.000565 0.001265 0.001679 \n", - "37 163840 179789 ... 0.015521 0.005221 0.011660 \n", - "38 163840 182577 ... -0.000140 0.001370 0.000755 \n", - "39 163840 182577 ... -0.032480 -0.033616 -0.029502 \n", - "40 163840 205369 ... -0.001281 -0.001583 -0.000117 \n", - "41 163840 205369 ... 0.006389 0.010391 0.008292 \n", - "42 163840 185065 ... 0.001077 0.002305 0.001138 \n", - "43 163840 185065 ... 0.009587 0.012155 0.002769 \n", - "44 163840 227488 ... 0.001978 0.001395 0.001501 \n", - "45 163840 227488 ... -0.004803 0.004060 -0.018828 \n", - "46 163840 174642 ... 0.000218 0.000532 0.000263 \n", - "47 163840 174642 ... -0.000386 -0.000480 -0.000415 \n", + "0 163840 172672 ... -0.038816 -0.132628 0.012945 \n", + "1 163840 178946 ... 0.007422 -0.040294 0.037184 \n", + "2 163840 170729 ... -0.038910 -0.132768 0.012589 \n", + "3 163840 165050 ... 0.006907 -0.041188 0.036539 \n", + "4 163840 166306 ... 0.384163 0.383833 0.208923 \n", + "5 163840 167179 ... 0.128392 0.201155 0.096084 \n", + "6 163840 198383 ... 0.382839 0.387868 0.207169 \n", + "7 163840 184493 ... 0.127002 0.199668 0.092870 \n", + "8 163840 168975 ... -0.065035 -0.009808 -0.089304 \n", + "9 163840 173223 ... 0.149715 0.036652 0.236477 \n", + "10 163840 170339 ... -0.064135 -0.008839 -0.088697 \n", + "11 163840 177242 ... 0.149233 0.036425 0.236163 \n", + "12 163840 166924 ... 0.025078 0.316090 0.052550 \n", + "13 163840 166653 ... -0.060781 0.136466 0.222455 \n", + "14 163840 199838 ... 0.024326 0.315336 0.050626 \n", + "15 163840 178999 ... -0.069126 0.131456 0.215256 \n", + "16 163840 168116 ... -0.068721 -0.131200 -0.095311 \n", + "17 163840 183816 ... -0.000722 -0.065477 -0.015448 \n", + "18 163840 181059 ... -0.068234 -0.130586 -0.095205 \n", + "19 163840 176946 ... -0.000299 -0.064919 -0.015113 \n", + "20 163840 177982 ... 0.144722 0.096918 0.053467 \n", + "21 163840 172928 ... -0.204408 -0.164630 -0.161219 \n", + "22 163840 180515 ... 0.145138 0.099315 0.055320 \n", + "23 163840 180515 ... -0.216383 -0.174637 -0.171691 \n", + "24 163840 179428 ... 0.125082 0.187445 0.094098 \n", + "25 163840 179428 ... -0.072443 -0.081550 -0.038486 \n", + "26 163840 173165 ... 0.125406 0.187519 0.094387 \n", + "27 163840 173165 ... -0.072282 -0.081517 -0.038386 \n", + "28 163840 172720 ... -0.109378 -0.095939 -0.046820 \n", + "29 163840 172720 ... -0.014295 -0.047262 0.040566 \n", + "30 163840 172374 ... -0.114771 -0.101728 -0.051283 \n", + "31 163840 172374 ... -0.014773 -0.051328 0.039865 \n", + "32 163840 170946 ... 0.017597 -0.080504 0.049995 \n", + "33 163840 170946 ... -0.297318 0.036640 0.039409 \n", + "34 163840 233549 ... 0.017213 -0.081016 0.048742 \n", + "35 163840 233549 ... -0.296310 0.027849 0.034702 \n", + "36 163840 179789 ... 0.173036 0.238832 -0.112437 \n", + "37 163840 179789 ... -0.587394 -0.383097 -0.074972 \n", + "38 163840 182577 ... 0.196702 0.244239 -0.096241 \n", + "39 163840 182577 ... -0.598326 -0.350383 -0.062013 \n", + "40 163840 205369 ... -0.065015 -0.134536 -0.140023 \n", + "41 163840 205369 ... -0.138112 -0.135524 -0.122887 \n", + "42 163840 185065 ... -0.064936 -0.134381 -0.140410 \n", + "43 163840 185065 ... -0.138833 -0.136745 -0.123534 \n", + "44 163840 227488 ... -0.260044 -0.168338 -0.271208 \n", + "45 163840 227488 ... 0.107979 0.020519 0.216325 \n", + "46 163840 174642 ... -0.256797 -0.163505 -0.266977 \n", + "47 163840 174642 ... 0.100775 0.018718 0.213268 \n", "\n", " agg_9531 agg_9532 agg_9533 agg_9535 agg_9536 agg_9537 agg_9538 \n", - "0 -0.000043 -0.000076 -0.000072 -0.000106 0.000029 -0.000055 -0.000119 \n", - "1 -0.000324 -0.000471 -0.000258 -0.000382 -0.000683 -0.000459 -0.000757 \n", - "2 -0.000442 -0.000589 -0.000313 -0.000526 -0.000856 -0.000637 -0.000881 \n", - "3 -0.001091 -0.001184 -0.000717 -0.001306 -0.001550 -0.001349 -0.001550 \n", - "4 -0.000051 0.000037 0.000028 -0.000024 0.000093 -0.000080 -0.000203 \n", - "5 -0.000328 -0.000361 -0.000211 -0.000395 -0.000428 -0.000371 -0.000468 \n", - "6 -0.000193 -0.000205 -0.000113 -0.000245 -0.000265 -0.000241 -0.000347 \n", - "7 0.000903 0.001198 0.001515 0.001150 0.001019 0.000498 0.001034 \n", - "8 -0.000307 -0.000214 -0.000158 -0.000330 -0.000363 -0.000348 -0.000509 \n", - "9 0.000009 -0.000056 -0.000199 -0.000207 -0.000112 0.000043 0.000051 \n", - "10 -0.000142 -0.000220 -0.000106 -0.000170 -0.000272 -0.000170 -0.000231 \n", - "11 0.007764 -0.001580 0.002468 0.008656 0.002925 0.007485 0.010022 \n", - "12 -0.000197 -0.000195 -0.000144 -0.000220 -0.000190 -0.000199 -0.000302 \n", - "13 -0.000158 -0.000164 -0.000123 -0.000158 -0.000281 -0.000191 -0.000319 \n", - "14 -0.001977 -0.000437 -0.000555 -0.001084 -0.000583 -0.001613 -0.000536 \n", - "15 -0.000099 -0.000132 -0.000082 -0.000092 -0.000189 -0.000112 -0.000195 \n", - "16 -0.000359 -0.000367 -0.000241 -0.000268 -0.000487 -0.000344 -0.000595 \n", - "17 -0.000674 -0.000717 -0.000515 -0.000520 -0.000780 -0.000614 -0.001010 \n", - "18 -0.000615 -0.000704 -0.000456 -0.000404 -0.000761 -0.000572 -0.000986 \n", - "19 0.000035 0.000096 0.000057 0.000009 0.000184 0.000039 0.000044 \n", - "20 -0.000260 -0.000322 -0.000207 -0.000179 -0.000336 -0.000235 -0.000382 \n", - "21 -0.000067 -0.000092 -0.000057 -0.000034 -0.000212 -0.000063 -0.000159 \n", - "22 0.000038 0.000027 0.000055 0.000078 0.000277 0.000062 0.000245 \n", - "23 0.001438 0.001558 0.001273 0.001937 0.005484 0.002190 0.005321 \n", - "24 0.001251 0.001342 0.000177 0.001402 0.000480 0.001663 0.001764 \n", - "25 -0.004169 -0.003774 -0.003399 -0.003382 0.000949 -0.002237 -0.002981 \n", - "26 -0.000085 0.000037 -0.000040 -0.000056 -0.000044 -0.000046 -0.000027 \n", - "27 0.001486 0.003162 0.001671 0.000299 0.002102 0.001007 0.001831 \n", - "28 0.000474 0.000076 0.000280 0.000422 -0.000033 0.000436 0.000369 \n", - "29 0.002841 0.000024 0.000726 0.001582 0.000497 0.002237 -0.000092 \n", - "30 0.000130 -0.000119 -0.000066 0.000112 0.000059 0.000113 0.000135 \n", - "31 0.000516 0.000702 0.000260 0.000547 0.000815 0.000190 0.000561 \n", - "32 -0.000305 -0.000203 -0.000045 -0.000183 -0.000423 -0.000412 -0.000544 \n", - "33 0.007996 0.015255 0.007950 0.010278 0.021851 0.009230 0.024402 \n", - "34 0.001515 -0.001879 -0.003638 0.004167 -0.000977 0.003568 0.001709 \n", - "35 0.007504 0.011276 0.019385 -0.002945 0.020405 -0.001569 0.007003 \n", - "36 -0.000034 -0.002686 -0.000878 0.001329 -0.001112 0.002059 0.000288 \n", - "37 0.028807 0.010368 0.014050 0.007810 0.026553 0.009914 0.026500 \n", - "38 0.000528 -0.000446 -0.000184 -0.001062 0.000593 0.000575 0.002940 \n", - "39 -0.050562 -0.030190 -0.021981 -0.012192 -0.029633 -0.037296 -0.067220 \n", - "40 -0.000501 -0.001837 -0.001011 -0.002222 0.000031 0.000068 0.000062 \n", - "41 0.010647 -0.005511 0.002638 -0.004853 -0.009829 0.008542 0.011590 \n", - "42 -0.000056 -0.000738 0.000059 0.000456 -0.000396 0.000904 0.000512 \n", - "43 0.018908 0.010602 0.000302 0.003231 0.007630 0.011715 0.019595 \n", - "44 0.000012 -0.000538 -0.000914 0.002817 -0.001998 0.001149 0.002069 \n", - "45 -0.000383 0.042374 0.022364 -0.023450 0.043353 0.006033 -0.006614 \n", - "46 0.000122 0.000352 0.000184 0.000139 0.000887 0.000173 0.000558 \n", - "47 -0.000092 -0.000513 -0.000241 -0.000285 -0.000141 -0.000277 -0.000758 \n", + "0 -0.057034 -0.045095 -0.019845 -0.048861 -0.248287 -0.157361 -0.202191 \n", + "1 -0.039102 -0.069273 -0.089758 0.085668 -0.100051 0.199948 -0.228135 \n", + "2 -0.057297 -0.044415 -0.019266 -0.049203 -0.247513 -0.158220 -0.202740 \n", + "3 -0.038984 -0.070001 -0.090065 0.085498 -0.100311 0.199196 -0.229159 \n", + "4 0.197689 0.147001 0.214184 0.429241 0.046892 0.232819 0.122424 \n", + "5 0.117298 -0.126285 -0.059694 0.184677 -0.009753 0.305165 -0.107976 \n", + "6 0.192711 0.148570 0.212126 0.429055 0.040387 0.231833 0.117729 \n", + "7 0.113858 -0.132513 -0.059923 0.180629 -0.008512 0.302857 -0.112850 \n", + "8 -0.123990 -0.237687 -0.145043 -0.027265 -0.172405 -0.209000 -0.169588 \n", + "9 -0.033082 -0.059123 -0.109544 -0.116550 -0.056748 0.238428 -0.066090 \n", + "10 -0.123552 -0.236969 -0.144438 -0.026115 -0.171792 -0.208494 -0.170012 \n", + "11 -0.033192 -0.059114 -0.109789 -0.116972 -0.057177 0.238166 -0.067322 \n", + "12 -0.128108 -0.025624 0.018586 0.070953 0.035494 0.035036 0.198598 \n", + "13 0.180580 -0.077465 0.059501 -0.152240 0.021208 0.388104 0.088033 \n", + "14 -0.121845 -0.024810 0.020802 0.065568 0.039766 0.038300 0.207067 \n", + "15 0.164175 -0.088568 0.049765 -0.160795 0.012626 0.377364 0.080307 \n", + "16 0.046062 0.058227 -0.014569 -0.110806 0.156508 0.050760 0.086102 \n", + "17 -0.048512 0.080478 0.044108 -0.040190 -0.109651 0.016822 -0.062620 \n", + "18 0.046179 0.059056 -0.013978 -0.110189 0.157281 0.050832 0.085974 \n", + "19 -0.048239 0.080818 0.044669 -0.040059 -0.109231 0.017035 -0.062525 \n", + "20 0.178367 0.186895 -0.086145 0.078297 0.339905 0.164013 0.351803 \n", + "21 -0.077524 0.001738 -0.079267 -0.283000 -0.061010 -0.083843 -0.061747 \n", + "22 0.177228 0.190383 -0.084195 0.083160 0.344412 0.166400 0.354686 \n", + "23 -0.086247 -0.017797 -0.098000 -0.294861 -0.081399 -0.096354 -0.078279 \n", + "24 0.073541 0.153524 0.122184 0.178186 0.325020 0.131711 0.251710 \n", + "25 -0.047714 0.226176 0.045261 -0.126760 0.371777 0.060598 -0.005164 \n", + "26 0.074029 0.154567 0.122893 0.178522 0.326261 0.131973 0.251846 \n", + "27 -0.047584 0.226377 0.045594 -0.126571 0.372053 0.060832 -0.004993 \n", + "28 -0.160233 -0.108172 -0.116325 0.049603 0.171065 0.111874 0.250584 \n", + "29 0.034287 0.086552 -0.002746 -0.036511 0.105897 0.075976 -0.094442 \n", + "30 -0.167664 -0.115162 -0.118221 0.045587 0.164287 0.105892 0.242462 \n", + "31 0.034642 0.079301 -0.005801 -0.037709 0.102587 0.075023 -0.093098 \n", + "32 -0.011808 -0.271627 -0.061634 -0.068092 -0.169892 -0.139209 -0.134266 \n", + "33 0.172793 0.131960 0.099102 -0.127269 0.248293 0.078142 0.244484 \n", + "34 -0.012799 -0.271882 -0.061404 -0.068229 -0.168541 -0.140811 -0.135465 \n", + "35 0.172035 0.133321 0.101695 -0.133212 0.247434 0.074210 0.243297 \n", + "36 -0.396939 -0.048346 -0.349238 0.178761 -0.541497 -0.432859 -0.115770 \n", + "37 0.116038 -0.106850 -0.536714 -0.289871 0.275144 -0.386951 -0.475129 \n", + "38 -0.383522 -0.053860 -0.346986 0.204916 -0.547306 -0.428087 -0.109396 \n", + "39 0.137351 -0.113740 -0.549226 -0.316263 0.282046 -0.367199 -0.457674 \n", + "40 0.024319 -0.060507 -0.108153 -0.149791 -0.038775 -0.076944 -0.118178 \n", + "41 0.003460 -0.095921 -0.142851 -0.133408 -0.044298 -0.079864 0.063350 \n", + "42 0.024382 -0.060211 -0.106950 -0.150407 -0.038308 -0.077518 -0.118339 \n", + "43 0.002528 -0.096342 -0.143056 -0.133687 -0.044767 -0.080880 0.062666 \n", + "44 -0.247191 -0.235198 -0.383056 -0.163678 -0.055487 -0.088683 -0.155701 \n", + "45 0.572342 0.147766 0.106553 0.096116 0.279162 0.162525 0.332815 \n", + "46 -0.247593 -0.233061 -0.386286 -0.155122 -0.059363 -0.081943 -0.156882 \n", + "47 0.574074 0.151917 0.110753 0.088787 0.286400 0.165354 0.329970 \n", "\n", "[48 rows x 8870 columns]" ] }, - "execution_count": 17, + "execution_count": 20, "metadata": {}, "output_type": "execute_result" } @@ -4247,19 +4712,9 @@ }, { "cell_type": "code", - "execution_count": 18, + "execution_count": 21, "metadata": {}, - "outputs": [ - { - "name": "stderr", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact decima_metadata:latest, 628.05MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:0.5 (1290.4MB/s)\n" - ] - } - ], + "outputs": [], "source": [ "from decima.data.dataset import VariantDataset\n", "\n", @@ -4275,7 +4730,7 @@ }, { "cell_type": "code", - "execution_count": 19, + "execution_count": 22, "metadata": {}, "outputs": [ { @@ -4284,7 +4739,7 @@ "164" ] }, - "execution_count": 19, + "execution_count": 22, "metadata": {}, "output_type": "execute_result" } @@ -4295,7 +4750,7 @@ }, { "cell_type": "code", - "execution_count": 20, + "execution_count": 23, "metadata": {}, "outputs": [ { @@ -4309,7 +4764,7 @@ " 'warning': []}" ] }, - "execution_count": 20, + "execution_count": 23, "metadata": {}, "output_type": "execute_result" } @@ -4320,7 +4775,7 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 24, "metadata": {}, "outputs": [ { @@ -4329,7 +4784,7 @@ "torch.Size([5, 524288])" ] }, - "execution_count": 21, + "execution_count": 24, "metadata": {}, "output_type": "execute_result" } @@ -4340,7 +4795,7 @@ }, { "cell_type": "code", - "execution_count": 22, + "execution_count": 25, "metadata": {}, "outputs": [ { @@ -4495,10 +4950,10 @@ " +\n", " 163840\n", " 227488\n", - " 180346\n", + " 180345\n", " TTT\n", " G\n", - " 16506\n", + " 16505\n", " \n", " \n", " 78\n", @@ -4512,10 +4967,10 @@ " +\n", " 163840\n", " 227488\n", - " 180867\n", + " 180866\n", " T\n", " GG\n", - " 17027\n", + " 17026\n", " \n", " \n", " 79\n", @@ -4594,8 +5049,8 @@ "3 163840 178946 120841 A G -42999 \n", "4 163840 170729 145915 C T -17925 \n", ".. ... ... ... ... ... ... \n", - "77 163840 227488 180346 TTT G 16506 \n", - "78 163840 227488 180867 T GG 17027 \n", + "77 163840 227488 180345 TTT G 16505 \n", + "78 163840 227488 180866 T GG 17026 \n", "79 163840 174642 507041 T G 343201 \n", "80 163840 174642 506226 AAA C 342386 \n", "81 163840 174642 505705 A CC 341865 \n", @@ -4603,7 +5058,7 @@ "[82 rows x 14 columns]" ] }, - "execution_count": 22, + "execution_count": 25, "metadata": {}, "output_type": "execute_result" } @@ -4616,27 +5071,14 @@ "cell_type": "markdown", "metadata": {}, "source": [ - "Let's load model" + "Let's load the ensemble model of all 4 Decima replicates:" ] }, { "cell_type": "code", - "execution_count": null, + "execution_count": 26, "metadata": {}, - "outputs": [ - { - "name": "stderr", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact decima_rep0:latest, 2155.88MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:1.2 (1808.4MB/s)\n", - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact human_state_dict_fold0:latest, 709.30MB. 1 files... \n", - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n", - "Done. 0:0:0.5 (1487.8MB/s)\n" - ] - } - ], + "outputs": [], "source": [ "from decima.hub import load_decima_model\n", "\n", @@ -4652,34 +5094,46 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 27, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ - "Using default `ModelCheckpoint`. Consider installing `litmodels` package to enable `LitModelCheckpoint` for automatic upload to the Lightning model registry.\n", + "/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/lightning_fabric/plugins/environments/slurm.py:204: PossibleUserWarning: The `srun` command is available on your system but is not used. HINT: If your intention is to run Lightning on SLURM, prepend your python command with `srun` like so: srun python /gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib ...\n", "GPU available: True (cuda), used: True\n", "TPU available: False, using: 0 TPU cores\n", - "HPU available: False, using: 0 HPUs\n", + "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n", + "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n", "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n", - "/home/celikm5/miniforge3/envs/decima/lib/python3.11/site-packages/pytorch_lightning/trainer/connectors/data_connector.py:425: PossibleUserWarning: The 'predict_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=95` in the `DataLoader` to improve performance.\n" + "/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/pytorch_lightning/trainer/connectors/data_connector.py:434: PossibleUserWarning: The 'predict_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=19` in the `DataLoader` to improve performance.\n" ] }, { "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "24aebc5e88744bd7b1361c1fde3bcd26", - "version_major": 2, - "version_minor": 0 - }, + "text/html": [ + "
/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/rich/live.py:260: UserWarning: \n",
+       "install \"ipywidgets\" for Jupyter support\n",
+       "
\n" + ], "text/plain": [ - "Predicting: | | 0/? [00:00\n" + ], + "text/plain": [] + }, + "metadata": {}, + "output_type": "display_data" } ], "source": [ @@ -4690,12 +5144,12 @@ "cell_type": "markdown", "metadata": {}, "source": [ - "The `preds` are predicted expression for `alt - ref` alleles for each variant:" + "The `preds` are the difference between `alt - ref` allele predictions for each variant-gene pair, i.e. the predicted log fold change in gene expression." ] }, { "cell_type": "code", - "execution_count": 25, + "execution_count": 28, "metadata": {}, "outputs": [ { @@ -4704,7 +5158,7 @@ "(82, 8856)" ] }, - "execution_count": 25, + "execution_count": 28, "metadata": {}, "output_type": "execute_result" } @@ -4715,28 +5169,28 @@ }, { "cell_type": "code", - "execution_count": 26, + "execution_count": 29, "metadata": {}, "outputs": [ { "data": { "text/plain": [ - "array([[-3.1028390e-03, -3.5525560e-03, -3.5931766e-03, ...,\n", - " -5.2702725e-03, -3.3913553e-03, -6.6891909e-03],\n", - " [ 1.1469424e-04, 1.7350912e-04, 8.1568956e-05, ...,\n", - " 1.3417006e-04, 1.1980534e-05, -2.7149916e-05],\n", - " [ 6.0230494e-05, -8.2150102e-05, -2.6747584e-05, ...,\n", - " -2.7078092e-03, -1.8020868e-03, -3.0442923e-03],\n", + "array([[-2.8905511e-02, -3.0458324e-02, -2.7693681e-02, ...,\n", + " -1.3502419e-02, 8.7248422e-03, 1.7261244e-03],\n", + " [-2.9755570e-04, -2.7861074e-04, -1.2444891e-04, ...,\n", + " -3.7071109e-04, 1.9611418e-04, -8.2259066e-05],\n", + " [ 5.6408532e-04, 1.1925958e-04, -4.3001957e-05, ...,\n", + " 2.8621778e-04, 7.6048309e-04, 2.2458541e-03],\n", " ...,\n", - " [ 5.8137625e-04, 6.4936280e-04, 4.9999356e-04, ...,\n", - " 8.4596872e-04, 1.6248971e-04, 5.3709745e-04],\n", - " [ 2.6370510e-03, 2.7481169e-03, 1.8703863e-03, ...,\n", - " 5.3109229e-04, -1.4962628e-04, -3.7675351e-04],\n", - " [-2.8469041e-03, -3.2258853e-03, -2.4763122e-03, ...,\n", - " -3.5528541e-03, -8.5747242e-04, -2.2805035e-03]], dtype=float32)" + " [-2.1314561e-02, -2.1281566e-02, -2.0599853e-02, ...,\n", + " -1.9413859e-02, -2.6478251e-02, -2.1747496e-02],\n", + " [ 2.5212765e-04, 4.7219824e-04, 3.2710657e-04, ...,\n", + " 3.1447969e-04, 4.4250442e-04, 5.5153947e-04],\n", + " [ 2.3546554e-03, 1.8186197e-03, 1.9949824e-03, ...,\n", + " -1.5318021e-03, 2.7205609e-03, -3.7190318e-04]], dtype=float32)" ] }, - "execution_count": 26, + "execution_count": 29, "metadata": {}, "output_type": "execute_result" } @@ -4747,16 +5201,17 @@ }, { "cell_type": "code", - "execution_count": 27, + "execution_count": 30, "metadata": {}, "outputs": [ { "data": { "text/plain": [ - "Counter({'allele_mismatch_with_reference_genome': 26, 'unknown': 0})" + "Counter({'allele_mismatch_with_reference_genome': tensor(13),\n", + " 'unknown': tensor(0)})" ] }, - "execution_count": 27, + "execution_count": 30, "metadata": {}, "output_type": "execute_result" } @@ -4774,7 +5229,7 @@ }, { "cell_type": "code", - "execution_count": 28, + "execution_count": 31, "metadata": {}, "outputs": [], "source": [ @@ -4784,7 +5239,7 @@ }, { "cell_type": "code", - "execution_count": 29, + "execution_count": 32, "metadata": {}, "outputs": [ { @@ -4793,7 +5248,7 @@ "torch.Size([2, 5, 524288])" ] }, - "execution_count": 29, + "execution_count": 32, "metadata": {}, "output_type": "execute_result" } @@ -4804,7 +5259,7 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 33, "metadata": {}, "outputs": [], "source": [ @@ -4838,7 +5293,7 @@ }, { "data": { - "image/png": "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", + "image/png": "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", "text/plain": [ "
" ] @@ -4859,9 +5314,9 @@ ], "metadata": { "kernelspec": { - "display_name": "Python 3 (ipykernel)", + "display_name": "Python (decima)", "language": "python", - "name": "python3" + "name": "decima" }, "language_info": { "codemirror_mode": { @@ -4873,7 +5328,7 @@ "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", - "version": "3.11.10" + "version": "3.11.15" } }, "nbformat": 4, diff --git a/docs/tutorials/5-gene-expression-prediction.ipynb b/docs/tutorials/5-gene-expression-prediction.ipynb index 9aa599c..e64d911 100644 --- a/docs/tutorials/5-gene-expression-prediction.ipynb +++ b/docs/tutorials/5-gene-expression-prediction.ipynb @@ -53,38 +53,8 @@ "name": "stderr", "output_type": "stream", "text": [ - "/home/celikm5/miniforge3/envs/decima2/lib/python3.11/site-packages/pydantic/_internal/_generate_schema.py:2249: UnsupportedFieldAttributeWarning: The 'repr' attribute with value False was provided to the `Field()` function, which has no effect in the context it was used. 'repr' is field-specific metadata, and can only be attached to a model field using `Annotated` metadata or by assignment. This may have happened because an `Annotated` type alias using the `type` statement was used, or if the `Field()` function was attached to a single member of a union type.\n", - " warnings.warn(\n", - "/home/celikm5/miniforge3/envs/decima2/lib/python3.11/site-packages/pydantic/_internal/_generate_schema.py:2249: UnsupportedFieldAttributeWarning: The 'frozen' attribute with value True was provided to the `Field()` function, which has no effect in the context it was used. 'frozen' is field-specific metadata, and can only be attached to a model field using `Annotated` metadata or by assignment. This may have happened because an `Annotated` type alias using the `type` statement was used, or if the `Field()` function was attached to a single member of a union type.\n", - " warnings.warn(\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Currently logged in as: \u001b[33mmhcelik\u001b[0m (\u001b[33mmhcw\u001b[0m) to \u001b[32mhttps://api.wandb.ai\u001b[0m. Use \u001b[1m`wandb login --relogin`\u001b[0m to force relogin\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact 'metadata:latest', 3122.32MB. 1 files...\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "Done. 00:00:08.3 (375.3MB/s)\n" + "/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n", + " from .autonotebook import tqdm as notebook_tqdm\n" ] }, { @@ -1299,265 +1269,17 @@ }, { "cell_type": "markdown", - "id": "345ff486", - "metadata": { - "vscode": { - "languageId": "plaintext" - } - }, - "source": [ - "### CLI API" - ] - }, - { - "cell_type": "markdown", - "id": "86e9f055", + "id": "de5bfe6d", "metadata": {}, "source": [ - "If you want to perform gene expression prediction again, rather than using the precomputed scores, you can use the Decima command-line interface (CLI) to generate new predictions for any set of genes you specify. For example, you can run the `decima predict-genes` command with the `--genes` argument to provide a comma-separated list of gene names (such as \"STRADA,ETV4,USP25\") if no gene provided it will perform expression predictions for all genes, select the prediction model with the `--model` option (for instance, \"ensemble\" or a specific replicate like \"0\"), and use `--save-replicates` to save predictions for each replicate. The `-o` flag lets you specify the output file path for the predictions in `.h5ad` format. " + "Remember, to see the metadata for the various output tracks:" ] }, { "cell_type": "code", "execution_count": 5, - "id": "6715257c", - "metadata": { - "execution": { - "iopub.execute_input": "2025-11-21T06:25:32.221652Z", - "iopub.status.busy": "2025-11-21T06:25:32.221116Z", - "iopub.status.idle": "2025-11-21T06:26:16.929651Z", - "shell.execute_reply": "2025-11-21T06:26:16.928993Z" - } - }, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "/home/celikm5/miniforge3/envs/decima2/lib/python3.11/site-packages/pydantic/_internal/_generate_schema.py:2249: UnsupportedFieldAttributeWarning: The 'repr' attribute with value False was provided to the `Field()` function, which has no effect in the context it was used. 'repr' is field-specific metadata, and can only be attached to a model field using `Annotated` metadata or by assignment. This may have happened because an `Annotated` type alias using the `type` statement was used, or if the `Field()` function was attached to a single member of a union type.\r\n", - " warnings.warn(\r\n", - "/home/celikm5/miniforge3/envs/decima2/lib/python3.11/site-packages/pydantic/_internal/_generate_schema.py:2249: UnsupportedFieldAttributeWarning: The 'frozen' attribute with value True was provided to the `Field()` function, which has no effect in the context it was used. 'frozen' is field-specific metadata, and can only be attached to a model field using `Annotated` metadata or by assignment. This may have happened because an `Annotated` type alias using the `type` statement was used, or if the `Field()` function was attached to a single member of a union type.\r\n", - " warnings.warn(\r\n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "decima - INFO - Using device: 0 and genome: hg38 for prediction.\r\n", - "decima - INFO - Loading model ensemble...\r\n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Currently logged in as: \u001b[33mmhcelik\u001b[0m (\u001b[33mmhcw\u001b[0m) to \u001b[32mhttps://api.wandb.ai\u001b[0m. Use \u001b[1m`wandb login --relogin`\u001b[0m to force relogin\r\n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact 'rep0:latest', 720.03MB. 1 files...\r\n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \r\n", - "Done. 00:00:02.6 (272.9MB/s)\r\n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact 'rep1:latest', 720.03MB. 1 files...\r\n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \r\n", - "Done. 00:00:01.8 (405.6MB/s)\r\n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact 'rep2:latest', 720.03MB. 1 files...\r\n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \r\n", - "Done. 00:00:02.0 (359.5MB/s)\r\n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact 'rep3:latest', 720.03MB. 1 files...\r\n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \r\n", - "Done. 00:00:01.8 (400.1MB/s)\r\n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "decima - INFO - Making predictions\r\n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact 'metadata:latest', 3122.32MB. 1 files...\r\n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \r\n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "Done. 00:00:01.9 (1645.5MB/s)\r\n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "/home/celikm5/miniforge3/envs/decima2/lib/python3.11/site-packages/torch/__init__.py:1617: UserWarning: Please use the new API settings to control TF32 behavior, such as torch.backends.cudnn.conv.fp32_precision = 'tf32' or torch.backends.cuda.matmul.fp32_precision = 'ieee'. Old settings, e.g, torch.backends.cuda.matmul.allow_tf32 = True, torch.backends.cudnn.allow_tf32 = True, allowTF32CuDNN() and allowTF32CuBLAS() will be deprecated after Pytorch 2.9. Please see https://pytorch.org/docs/main/notes/cuda.html#tensorfloat-32-tf32-on-ampere-and-later-devices (Triggered internally at /pytorch/aten/src/ATen/Context.cpp:80.)\r\n", - "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\r\n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "GPU available: True (cuda), used: True\r\n", - "TPU available: False, using: 0 TPU cores\r\n", - "HPU available: False, using: 0 HPUs\r\n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\r\n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "SLURM auto-requeueing enabled. Setting signal handlers.\r\n" - ] - }, - { - "name": "stdout", - "output_type": "stream", - "text": [ - "\r", - "Predicting: | | 0/? [00:00\n", " \n", " \n", - " STRADA\n", - " ETV4\n", - " USP25\n", + " cell_type\n", + " tissue\n", + " organ\n", + " disease\n", + " study\n", + " dataset\n", + " region\n", + " subregion\n", + " celltype_coarse\n", + " n_cells\n", + " total_counts\n", + " n_genes\n", + " size_factor\n", + " train_pearson\n", + " val_pearson\n", + " test_pearson\n", " \n", " \n", " \n", " \n", " agg_0\n", - " 3.060561\n", - " 2.882100\n", - " 3.469085\n", + " Amygdala excitatory\n", + " Amygdala_Amygdala\n", + " CNS\n", + " healthy\n", + " jhpce#tran2021\n", + " brain_atlas\n", + " Amygdala\n", + " Amygdala\n", + " NaN\n", + " 331\n", + " 1.592883e+07\n", + " 17000\n", + " 41431.465186\n", + " 0.942459\n", + " 0.841377\n", + " 0.865640\n", " \n", " \n", " agg_1\n", - " 3.117702\n", - " 2.825738\n", - " 3.596372\n", + " Amygdala excitatory\n", + " Amygdala_Basolateral nuclear group (BLN) - lat...\n", + " CNS\n", + " healthy\n", + " SCR_016152\n", + " brain_atlas\n", + " Amygdala\n", + " Basolateral nuclear group (BLN) - lateral nucl...\n", + " NaN\n", + " 11369\n", + " 2.952133e+08\n", + " 18080\n", + " 40765.341481\n", + " 0.943098\n", + " 0.838936\n", + " 0.861092\n", " \n", " \n", " agg_2\n", - " 3.156642\n", - " 3.122381\n", - " 3.718749\n", + " Amygdala excitatory\n", + " Amygdala_Bed nucleus of stria terminalis and n...\n", + " CNS\n", + " healthy\n", + " SCR_016152\n", + " brain_atlas\n", + " Amygdala\n", + " Bed nucleus of stria terminalis and nearby - BNST\n", + " NaN\n", + " 139\n", + " 2.593231e+06\n", + " 15418\n", + " 42556.387020\n", + " 0.952170\n", + " 0.854544\n", + " 0.866654\n", " \n", " \n", " agg_3\n", - " 3.214047\n", - " 3.204670\n", - " 3.629874\n", + " Amygdala excitatory\n", + " Amygdala_Central nuclear group - CEN\n", + " CNS\n", + " healthy\n", + " SCR_016152\n", + " brain_atlas\n", + " Amygdala\n", + " Central nuclear group - CEN\n", + " NaN\n", + " 3892\n", + " 9.946371e+07\n", + " 17959\n", + " 42884.641430\n", + " 0.959744\n", + " 0.863585\n", + " 0.881554\n", " \n", " \n", " agg_4\n", - " 3.103570\n", - " 3.032028\n", - " 3.512117\n", + " Amygdala excitatory\n", + " Amygdala_Corticomedial nuclear group (CMN) - a...\n", + " CNS\n", + " healthy\n", + " SCR_016152\n", + " brain_atlas\n", + " Amygdala\n", + " Corticomedial nuclear group (CMN) - anterior c...\n", + " NaN\n", + " 2945\n", + " 1.281619e+08\n", + " 17885\n", + " 41816.741933\n", + " 0.951365\n", + " 0.854304\n", + " 0.868902\n", " \n", " \n", " ...\n", " ...\n", " ...\n", " ...\n", + " ...\n", + " ...\n", + " ...\n", + " ...\n", + " ...\n", + " ...\n", + " ...\n", + " ...\n", + " ...\n", + " ...\n", + " ...\n", + " ...\n", + " ...\n", " \n", " \n", " agg_9533\n", - " 2.313220\n", - " 2.173597\n", - " 3.156157\n", + " vascular associated smooth muscle cell\n", + " upper lobe of right lung\n", + " lung\n", + " NA\n", + " ENCODE\n", + " scimilarity\n", + " nan\n", + " nan\n", + " NaN\n", + " 21\n", + " 3.483375e+04\n", + " 8515\n", + " 35404.911768\n", + " 0.735213\n", + " 0.665647\n", + " 0.654491\n", " \n", " \n", " agg_9535\n", - " 0.952112\n", - " 0.956720\n", - " 1.250005\n", + " vascular associated smooth muscle cell\n", + " urinary bladder\n", + " urinary\n", + " healthy\n", + " GSE129845\n", + " scimilarity\n", + " nan\n", + " nan\n", + " NaN\n", + " 24\n", + " 8.498500e+04\n", + " 7337\n", + " 26189.415789\n", + " 0.809852\n", + " 0.690022\n", + " 0.656160\n", " \n", " \n", " agg_9536\n", - " 2.779494\n", - " 2.705945\n", - " 3.530610\n", + " vascular associated smooth muscle cell\n", + " uterus\n", + " uterus\n", + " NA\n", + " ENCODE\n", + " scimilarity\n", + " nan\n", + " nan\n", + " NaN\n", + " 272\n", + " 5.700762e+05\n", + " 14769\n", + " 44938.403867\n", + " 0.915329\n", + " 0.808941\n", + " 0.839993\n", " \n", " \n", " agg_9537\n", - " 1.342694\n", - " 1.407407\n", - " 1.867070\n", + " vascular associated smooth muscle cell\n", + " uterus\n", + " uterus\n", + " healthy\n", + " e5f58829-1a66-40b5-a624-9046778e74f5\n", + " scimilarity\n", + " nan\n", + " nan\n", + " NaN\n", + " 472\n", + " 1.089170e+07\n", + " 14514\n", + " 30145.422152\n", + " 0.852339\n", + " 0.717682\n", + " 0.727469\n", " \n", " \n", " agg_9538\n", - " 1.745080\n", - " 1.633314\n", - " 2.082985\n", + " vascular associated smooth muscle cell\n", + " vasculature\n", + " vasculature\n", + " healthy\n", + " e5f58829-1a66-40b5-a624-9046778e74f5\n", + " scimilarity\n", + " nan\n", + " nan\n", + " NaN\n", + " 1853\n", + " 5.992697e+07\n", + " 16764\n", + " 36464.273371\n", + " 0.909855\n", + " 0.780413\n", + " 0.796351\n", " \n", " \n", "\n", - "

8856 rows × 3 columns

\n", + "

8856 rows × 16 columns

\n", "" ], "text/plain": [ - " STRADA ETV4 USP25\n", - "agg_0 3.060561 2.882100 3.469085\n", - "agg_1 3.117702 2.825738 3.596372\n", - "agg_2 3.156642 3.122381 3.718749\n", - "agg_3 3.214047 3.204670 3.629874\n", - "agg_4 3.103570 3.032028 3.512117\n", - "... ... ... ...\n", - "agg_9533 2.313220 2.173597 3.156157\n", - "agg_9535 0.952112 0.956720 1.250005\n", - "agg_9536 2.779494 2.705945 3.530610\n", - "agg_9537 1.342694 1.407407 1.867070\n", - "agg_9538 1.745080 1.633314 2.082985\n", + " cell_type \\\n", + "agg_0 Amygdala excitatory \n", + "agg_1 Amygdala excitatory \n", + "agg_2 Amygdala excitatory \n", + "agg_3 Amygdala excitatory \n", + "agg_4 Amygdala excitatory \n", + "... ... \n", + "agg_9533 vascular associated smooth muscle cell \n", + "agg_9535 vascular associated smooth muscle cell \n", + "agg_9536 vascular associated smooth muscle cell \n", + "agg_9537 vascular associated smooth muscle cell \n", + "agg_9538 vascular associated smooth muscle cell \n", "\n", - "[8856 rows x 3 columns]" + " tissue organ \\\n", + "agg_0 Amygdala_Amygdala CNS \n", + "agg_1 Amygdala_Basolateral nuclear group (BLN) - lat... CNS \n", + "agg_2 Amygdala_Bed nucleus of stria terminalis and n... CNS \n", + "agg_3 Amygdala_Central nuclear group - CEN CNS \n", + "agg_4 Amygdala_Corticomedial nuclear group (CMN) - a... CNS \n", + "... ... ... \n", + "agg_9533 upper lobe of right lung lung \n", + "agg_9535 urinary bladder urinary \n", + "agg_9536 uterus uterus \n", + "agg_9537 uterus uterus \n", + "agg_9538 vasculature vasculature \n", + "\n", + " disease study dataset \\\n", + "agg_0 healthy jhpce#tran2021 brain_atlas \n", + "agg_1 healthy SCR_016152 brain_atlas \n", + "agg_2 healthy SCR_016152 brain_atlas \n", + "agg_3 healthy SCR_016152 brain_atlas \n", + "agg_4 healthy SCR_016152 brain_atlas \n", + "... ... ... ... \n", + "agg_9533 NA ENCODE scimilarity \n", + "agg_9535 healthy GSE129845 scimilarity \n", + "agg_9536 NA ENCODE scimilarity \n", + "agg_9537 healthy e5f58829-1a66-40b5-a624-9046778e74f5 scimilarity \n", + "agg_9538 healthy e5f58829-1a66-40b5-a624-9046778e74f5 scimilarity \n", + "\n", + " region subregion \\\n", + "agg_0 Amygdala Amygdala \n", + "agg_1 Amygdala Basolateral nuclear group (BLN) - lateral nucl... \n", + "agg_2 Amygdala Bed nucleus of stria terminalis and nearby - BNST \n", + "agg_3 Amygdala Central nuclear group - CEN \n", + "agg_4 Amygdala Corticomedial nuclear group (CMN) - anterior c... \n", + "... ... ... \n", + "agg_9533 nan nan \n", + "agg_9535 nan nan \n", + "agg_9536 nan nan \n", + "agg_9537 nan nan \n", + "agg_9538 nan nan \n", + "\n", + " celltype_coarse n_cells total_counts n_genes size_factor \\\n", + "agg_0 NaN 331 1.592883e+07 17000 41431.465186 \n", + "agg_1 NaN 11369 2.952133e+08 18080 40765.341481 \n", + "agg_2 NaN 139 2.593231e+06 15418 42556.387020 \n", + "agg_3 NaN 3892 9.946371e+07 17959 42884.641430 \n", + "agg_4 NaN 2945 1.281619e+08 17885 41816.741933 \n", + "... ... ... ... ... ... \n", + "agg_9533 NaN 21 3.483375e+04 8515 35404.911768 \n", + "agg_9535 NaN 24 8.498500e+04 7337 26189.415789 \n", + "agg_9536 NaN 272 5.700762e+05 14769 44938.403867 \n", + "agg_9537 NaN 472 1.089170e+07 14514 30145.422152 \n", + "agg_9538 NaN 1853 5.992697e+07 16764 36464.273371 \n", + "\n", + " train_pearson val_pearson test_pearson \n", + "agg_0 0.942459 0.841377 0.865640 \n", + "agg_1 0.943098 0.838936 0.861092 \n", + "agg_2 0.952170 0.854544 0.866654 \n", + "agg_3 0.959744 0.863585 0.881554 \n", + "agg_4 0.951365 0.854304 0.868902 \n", + "... ... ... ... \n", + "agg_9533 0.735213 0.665647 0.654491 \n", + "agg_9535 0.809852 0.690022 0.656160 \n", + "agg_9536 0.915329 0.808941 0.839993 \n", + "agg_9537 0.852339 0.717682 0.727469 \n", + "agg_9538 0.909855 0.780413 0.796351 \n", + "\n", + "[8856 rows x 16 columns]" ] }, - "execution_count": 6, + "execution_count": 5, "metadata": {}, "output_type": "execute_result" } ], "source": [ - "result = DecimaResult.load(\"example/predict_genes/predictions.h5ad\")\n", - "result.predicted_expression_matrix()" + "result.cell_metadata" + ] + }, + { + "cell_type": "markdown", + "id": "345ff486", + "metadata": { + "vscode": { + "languageId": "plaintext" + } + }, + "source": [ + "### CLI API" + ] + }, + { + "cell_type": "markdown", + "id": "86e9f055", + "metadata": {}, + "source": [ + "If you want to perform gene expression prediction again, rather than using the precomputed scores, you can use the Decima command-line interface (CLI) to generate new predictions for any set of genes you specify. For example, you can run the `decima predict-genes` command with the `--genes` argument to provide a comma-separated list of gene names (such as \"STRADA,ETV4,USP25\") if no gene provided it will perform expression predictions for all genes, select the prediction model with the `--model` option (for instance, \"ensemble\" or a specific replicate like \"0\"), and use `--save-replicates` to save predictions for each replicate. The `-o` flag lets you specify the output file path for the predictions in `.h5ad` format. " + ] + }, + { + "cell_type": "code", + "execution_count": 6, + "id": "6715257c", + "metadata": { + "execution": { + "iopub.execute_input": "2025-11-21T06:25:32.221652Z", + "iopub.status.busy": "2025-11-21T06:25:32.221116Z", + "iopub.status.idle": "2025-11-21T06:26:16.929651Z", + "shell.execute_reply": "2025-11-21T06:26:16.928993Z" + } + }, + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "decima - INFO - Using device: 0 and genome: hg38 for prediction.\n", + "decima - INFO - Loading model ensemble...\n", + "Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.\n", + "decima - INFO - Making predictions\n", + "/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/lightning_fabric/plugins/environments/slurm.py:204: PossibleUserWarning: The `srun` command is available on your system but is not used. HINT: If your intention is to run Lightning on SLURM, prepend your python command with `srun` like so: srun python /gpfs/scratchfs01/site/u/lala8/conda/envs/decima/bin ...\n", + "GPU available: True (cuda), used: True\n", + "TPU available: False, using: 0 TPU cores\n", + "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n", + "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n", + "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n", + "\u001b[2KPredicting \u001b[35m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m 3/3 \u001b[2m0:00:06 • 0:00:00\u001b[0m \u001b[2;4m0.51it/s\u001b[0m [2;4m0.51it/s\u001b[0m \n", + "\u001b[?25h/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/torchmetrics/utilities/prints.py:43: UserWarning: The ``compute`` method of metric WarningCounter was called before the ``update`` method which may lead to errors, as metric states have not yet been updated.\n", + "decima - INFO - Creating anndata\n", + "decima - INFO - Evaluating performance\n", + "decima - WARNING - No ground truth expression matrix found in the metadata. Skipping evaluation.\n", + "\u001b[0m" + ] + } + ], + "source": [ + "! decima predict-genes --genes \"STRADA,ETV4,USP25\" --model ensemble --save-replicates -o example/predict_genes/predictions.h5ad " + ] + }, + { + "cell_type": "markdown", + "id": "4d2c0356", + "metadata": {}, + "source": [ + "After running this command, you can load the resulting predictions in Python using the `DecimaResult` class and access the predicted expression matrix as:" + ] + }, + { + "cell_type": "code", + "execution_count": 7, + "id": "e598b85c", + "metadata": { + "execution": { + "iopub.execute_input": "2025-11-21T06:26:16.931282Z", + "iopub.status.busy": "2025-11-21T06:26:16.931107Z", + "iopub.status.idle": "2025-11-21T06:26:16.979662Z", + "shell.execute_reply": "2025-11-21T06:26:16.979173Z" + } + }, + "outputs": [ + { + "data": { + "text/html": [ + "
\n", + "\n", + "\n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + " \n", + "
STRADAETV4USP25
agg_03.0606192.8825313.468806
agg_13.1177082.8263383.596656
agg_23.1566163.1228823.718929
agg_33.2139843.2050033.629975
agg_43.1033433.0324273.511971
............
agg_95332.3136262.1737693.155950
agg_95350.9520150.9565261.249992
agg_95362.7794072.7058983.530358
agg_95371.3424511.4072461.866810
agg_95381.7448641.6331262.082733
\n", + "

8856 rows × 3 columns

\n", + "
" + ], + "text/plain": [ + " STRADA ETV4 USP25\n", + "agg_0 3.060619 2.882531 3.468806\n", + "agg_1 3.117708 2.826338 3.596656\n", + "agg_2 3.156616 3.122882 3.718929\n", + "agg_3 3.213984 3.205003 3.629975\n", + "agg_4 3.103343 3.032427 3.511971\n", + "... ... ... ...\n", + "agg_9533 2.313626 2.173769 3.155950\n", + "agg_9535 0.952015 0.956526 1.249992\n", + "agg_9536 2.779407 2.705898 3.530358\n", + "agg_9537 1.342451 1.407246 1.866810\n", + "agg_9538 1.744864 1.633126 2.082733\n", + "\n", + "[8856 rows x 3 columns]" + ] + }, + "execution_count": 7, + "metadata": {}, + "output_type": "execute_result" + } + ], + "source": [ + "result = DecimaResult.load(\"example/predict_genes/predictions.h5ad\")\n", + "result.predicted_expression_matrix()" ] }, { @@ -1694,7 +1842,7 @@ }, { "cell_type": "code", - "execution_count": 7, + "execution_count": 8, "id": "ce3884f4", "metadata": { "execution": { @@ -1734,33 +1882,33 @@ " \n", " \n", " agg_0\n", - " 2.933017\n", - " 2.892934\n", - " 2.858879\n", + " 2.932055\n", + " 2.893697\n", + " 2.858597\n", " \n", " \n", " agg_1\n", - " 2.816778\n", - " 2.812964\n", - " 3.058485\n", + " 2.816373\n", + " 2.813228\n", + " 3.059142\n", " \n", " \n", " agg_2\n", - " 2.743120\n", - " 2.971366\n", - " 2.950323\n", + " 2.742780\n", + " 2.971545\n", + " 2.950576\n", " \n", " \n", " agg_3\n", - " 2.804692\n", - " 3.346689\n", - " 2.837382\n", + " 2.804382\n", + " 3.346670\n", + " 2.837498\n", " \n", " \n", " agg_4\n", - " 2.816030\n", - " 3.088620\n", - " 2.973081\n", + " 2.815310\n", + " 3.088664\n", + " 2.973327\n", " \n", " \n", " ...\n", @@ -1770,33 +1918,33 @@ " \n", " \n", " agg_9533\n", - " 2.510605\n", - " 2.760446\n", - " 1.965776\n", + " 2.510626\n", + " 2.761035\n", + " 1.965437\n", " \n", " \n", " agg_9535\n", - " 1.246022\n", - " 1.192524\n", - " 0.407104\n", + " 1.245751\n", + " 1.192484\n", + " 0.406721\n", " \n", " \n", " agg_9536\n", - " 2.809229\n", - " 3.369957\n", - " 2.927355\n", + " 2.808623\n", + " 3.369797\n", + " 2.927248\n", " \n", " \n", " agg_9537\n", - " 1.580334\n", - " 1.533467\n", - " 0.879179\n", + " 1.580036\n", + " 1.532877\n", + " 0.878610\n", " \n", " \n", " agg_9538\n", - " 2.014224\n", - " 2.078125\n", - " 1.374325\n", + " 2.013807\n", + " 2.077840\n", + " 1.373636\n", " \n", " \n", "\n", @@ -1805,22 +1953,22 @@ ], "text/plain": [ " STRADA ETV4 USP25\n", - "agg_0 2.933017 2.892934 2.858879\n", - "agg_1 2.816778 2.812964 3.058485\n", - "agg_2 2.743120 2.971366 2.950323\n", - "agg_3 2.804692 3.346689 2.837382\n", - "agg_4 2.816030 3.088620 2.973081\n", + "agg_0 2.932055 2.893697 2.858597\n", + "agg_1 2.816373 2.813228 3.059142\n", + "agg_2 2.742780 2.971545 2.950576\n", + "agg_3 2.804382 3.346670 2.837498\n", + "agg_4 2.815310 3.088664 2.973327\n", "... ... ... ...\n", - "agg_9533 2.510605 2.760446 1.965776\n", - "agg_9535 1.246022 1.192524 0.407104\n", - "agg_9536 2.809229 3.369957 2.927355\n", - "agg_9537 1.580334 1.533467 0.879179\n", - "agg_9538 2.014224 2.078125 1.374325\n", + "agg_9533 2.510626 2.761035 1.965437\n", + "agg_9535 1.245751 1.192484 0.406721\n", + "agg_9536 2.808623 3.369797 2.927248\n", + "agg_9537 1.580036 1.532877 0.878610\n", + "agg_9538 2.013807 2.077840 1.373636\n", "\n", "[8856 rows x 3 columns]" ] }, - "execution_count": 7, + "execution_count": 8, "metadata": {}, "output_type": "execute_result" } @@ -1851,7 +1999,7 @@ }, { "cell_type": "code", - "execution_count": 8, + "execution_count": 9, "id": "3a0dba6e", "metadata": { "execution": { @@ -1866,169 +2014,45 @@ "name": "stderr", "output_type": "stream", "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact 'rep0:latest', 720.03MB. 1 files...\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "Done. 00:00:01.3 (539.8MB/s)\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact 'rep1:latest', 720.03MB. 1 files...\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "Done. 00:00:01.4 (529.6MB/s)\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact 'rep2:latest', 720.03MB. 1 files...\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "Done. 00:00:00.6 (1166.2MB/s)\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact 'rep3:latest', 720.03MB. 1 files...\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "Done. 00:00:00.6 (1242.4MB/s)\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact 'metadata:latest', 3122.32MB. 1 files...\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "Done. 00:00:05.1 (613.2MB/s)\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "/home/celikm5/miniforge3/envs/decima2/lib/python3.11/site-packages/torch/__init__.py:1617: UserWarning: Please use the new API settings to control TF32 behavior, such as torch.backends.cudnn.conv.fp32_precision = 'tf32' or torch.backends.cuda.matmul.fp32_precision = 'ieee'. Old settings, e.g, torch.backends.cuda.matmul.allow_tf32 = True, torch.backends.cudnn.allow_tf32 = True, allowTF32CuDNN() and allowTF32CuBLAS() will be deprecated after Pytorch 2.9. Please see https://pytorch.org/docs/main/notes/cuda.html#tensorfloat-32-tf32-on-ampere-and-later-devices (Triggered internally at /pytorch/aten/src/ATen/Context.cpp:80.)\n", - "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "GPU available: True (cuda), used: True\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "TPU available: False, using: 0 TPU cores\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "HPU available: False, using: 0 HPUs\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ + "Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.\n", + "/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/lightning_fabric/plugins/environments/slurm.py:204: PossibleUserWarning: The `srun` command is available on your system but is not used. HINT: If your intention is to run Lightning on SLURM, prepend your python command with `srun` like so: srun python /gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib ...\n", + "GPU available: True (cuda), used: True\n", + "TPU available: False, using: 0 TPU cores\n", + "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n", + "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n", "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n" ] }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "SLURM auto-requeueing enabled. Setting signal handlers.\n" - ] - }, { "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "ff694bb6902a480b87789da9b1bb0156", - "version_major": 2, - "version_minor": 0 - }, + "text/html": [ + "
/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/rich/live.py:260: UserWarning: \n",
+       "install \"ipywidgets\" for Jupyter support\n",
+       "
\n" + ], "text/plain": [ - "Predicting: | | 0/? [00:00\n" + ], + "text/plain": [] + }, + "metadata": {}, + "output_type": "display_data" + }, { "name": "stderr", "output_type": "stream", "text": [ - "/home/celikm5/miniforge3/envs/decima2/lib/python3.11/site-packages/torchmetrics/utilities/prints.py:43: UserWarning: The ``compute`` method of metric WarningCounter was called before the ``update`` method which may lead to errors, as metric states have not yet been updated.\n", + "/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/torchmetrics/utilities/prints.py:43: UserWarning: The ``compute`` method of metric WarningCounter was called before the ``update`` method which may lead to errors, as metric states have not yet been updated.\n", "No ground truth expression matrix found in the metadata. Skipping evaluation.\n" ] } @@ -2062,7 +2086,7 @@ }, { "cell_type": "code", - "execution_count": 9, + "execution_count": 10, "id": "d147aaff", "metadata": { "execution": { @@ -2077,117 +2101,50 @@ "name": "stderr", "output_type": "stream", "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact 'rep0:latest', 720.03MB. 1 files...\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "Done. 00:00:01.2 (580.7MB/s)\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: Downloading large artifact 'metadata:latest', 3122.32MB. 1 files...\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "\u001b[34m\u001b[1mwandb\u001b[0m: 1 of 1 files downloaded. \n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "Done. 00:00:02.7 (1164.2MB/s)\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "GPU available: True (cuda), used: True\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "TPU available: False, using: 0 TPU cores\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "HPU available: False, using: 0 HPUs\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "SLURM auto-requeueing enabled. Setting signal handlers.\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "/home/celikm5/miniforge3/envs/decima2/lib/python3.11/site-packages/pytorch_lightning/trainer/connectors/data_connector.py:433: PossibleUserWarning: The 'predict_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=3` in the `DataLoader` to improve performance.\n" + "/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/lightning_fabric/plugins/environments/slurm.py:204: PossibleUserWarning: The `srun` command is available on your system but is not used. HINT: If your intention is to run Lightning on SLURM, prepend your python command with `srun` like so: srun python /gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib ...\n", + "GPU available: True (cuda), used: True\n", + "TPU available: False, using: 0 TPU cores\n", + "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n", + "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n", + "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n", + "/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/pytorch_lightning/trainer/connectors/data_connector.py:434: PossibleUserWarning: The 'predict_dataloader' does not have many workers which may be a bottleneck. Consider increasing the value of the `num_workers` argument` to `num_workers=19` in the `DataLoader` to improve performance.\n" ] }, { "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "086c3698b2f64da9b6be5915347e0fb7", - "version_major": 2, - "version_minor": 0 - }, + "text/html": [ + "
/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/rich/live.py:260: UserWarning: \n",
+       "install \"ipywidgets\" for Jupyter support\n",
+       "
\n" + ], "text/plain": [ - "Predicting: | | 0/? [00:00\n" + ], + "text/plain": [] + }, + "metadata": {}, + "output_type": "display_data" + }, { "name": "stdout", "output_type": "stream", "text": [ - "{'expression': array([[2.9330175, 2.8167777, 2.7431197, ..., 2.8092291, 1.580334 ,\n", - " 2.0142236],\n", - " [2.0205972, 1.9390993, 2.3948462, ..., 1.5032245, 0.51865 ,\n", - " 0.8084921],\n", - " [4.795624 , 4.9177237, 4.8060527, ..., 4.6602893, 2.7781048,\n", - " 3.1485069]], dtype=float32),\n", + "{'expression': array([[2.9320552, 2.8163733, 2.7427804, ..., 2.8086236, 1.580037 ,\n", + " 2.013807 ],\n", + " [2.0220397, 1.9408561, 2.396594 , ..., 1.5032645, 0.5187098,\n", + " 0.8085415],\n", + " [4.7951455, 4.9170322, 4.8060875, ..., 4.659208 , 2.77756 ,\n", + " 3.1479335]], dtype=float32),\n", " 'warnings': Counter({'unknown': tensor(0),\n", " 'allele_mismatch_with_reference_genome': tensor(0)})}\n" ] @@ -2196,7 +2153,7 @@ "name": "stderr", "output_type": "stream", "text": [ - "/home/celikm5/miniforge3/envs/decima2/lib/python3.11/site-packages/torchmetrics/utilities/prints.py:43: UserWarning: The ``compute`` method of metric WarningCounter was called before the ``update`` method which may lead to errors, as metric states have not yet been updated.\n" + "/gpfs/scratchfs01/site/u/lala8/conda/envs/decima/lib/python3.11/site-packages/torchmetrics/utilities/prints.py:43: UserWarning: The ``compute`` method of metric WarningCounter was called before the ``update`` method which may lead to errors, as metric states have not yet been updated.\n" ] } ], @@ -2242,7 +2199,7 @@ }, { "cell_type": "code", - "execution_count": 10, + "execution_count": 11, "id": "2dfbdb1a", "metadata": { "execution": { @@ -2257,7 +2214,7 @@ "name": "stdout", "output_type": "stream", "text": [ - "cat: ../tests/data/seqs.fasta: No such file or directory\r\n" + "cat: ../tests/data/seqs.fasta: No such file or directory\n" ] } ], @@ -2267,7 +2224,7 @@ }, { "cell_type": "code", - "execution_count": 11, + "execution_count": 12, "id": "0d15ea9e", "metadata": { "execution": { @@ -2282,54 +2239,19 @@ "name": "stderr", "output_type": "stream", "text": [ - "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "GPU available: True (cuda), used: True\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "TPU available: False, using: 0 TPU cores\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "HPU available: False, using: 0 HPUs\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ + "GPU available: True (cuda), used: True\n", + "TPU available: False, using: 0 TPU cores\n", + "💡 Tip: For seamless cloud logging and experiment tracking, try installing [litlogger](https://pypi.org/project/litlogger/) to enable LitLogger, which logs metrics and artifacts automatically to the Lightning Experiments platform.\n", + "💡 Tip: For seamless cloud uploads and versioning, try installing [litmodels](https://pypi.org/project/litmodels/) to enable LitModelCheckpoint, which syncs automatically with the Lightning model registry.\n", "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]\n" ] }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "SLURM auto-requeueing enabled. Setting signal handlers.\n" - ] - }, { "data": { - "application/vnd.jupyter.widget-view+json": { - "model_id": "b0c2854b5b1145a28d069646a54109b5", - "version_major": 2, - "version_minor": 0 - }, - "text/plain": [ - "Predicting: | | 0/? [00:00\n" + ], + "text/plain": [] }, "metadata": {}, "output_type": "display_data" @@ -2338,10 +2260,10 @@ "name": "stdout", "output_type": "stream", "text": [ - "array([[0.22026505, 0.21946757, 0.23086782, ..., 1.0358492 , 1.2124686 ,\n", - " 1.441437 ],\n", - " [0.26515013, 0.13341773, 0.14887223, ..., 0.44430685, 0.26546437,\n", - " 0.3042384 ]], dtype=float32)\n" + "array([[0.22032936, 0.21946222, 0.23088504, ..., 1.0364846 , 1.2134477 ,\n", + " 1.442827 ],\n", + " [0.26523364, 0.13350041, 0.14896177, ..., 0.44437748, 0.26560792,\n", + " 0.30437425]], dtype=float32)\n" ] } ], @@ -2366,9 +2288,9 @@ ], "metadata": { "kernelspec": { - "display_name": "decima2", + "display_name": "Python (decima)", "language": "python", - "name": "python3" + "name": "decima" }, "language_info": { "codemirror_mode": { @@ -2380,7 +2302,7 @@ "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", - "version": "3.11.14" + "version": "3.11.15" }, "widgets": { "application/vnd.jupyter.widget-state+json": { @@ -2924,13 +2846,13 @@ "description": "", "description_allow_html": false, "layout": "IPY_MODEL_15a6d137d2b646a696f272e5052ef607", - "max": 1.0, - "min": 0.0, + "max": 1, + "min": 0, "orientation": "horizontal", "style": "IPY_MODEL_aa835afeee484927b37b60145e71a78d", "tabbable": null, "tooltip": null, - "value": 1.0 + "value": 1 } }, "865789d5e3c44252bbf06d1074c3b9f4": { @@ -3003,13 +2925,13 @@ "description": "", "description_allow_html": false, "layout": "IPY_MODEL_90b1a10420424e418ff34bf3dd4412cf", - "max": 1.0, - "min": 0.0, + "max": 1, + "min": 0, "orientation": "horizontal", "style": "IPY_MODEL_f8201ba403ce4b2595d880cc13f891ad", "tabbable": null, "tooltip": null, - "value": 1.0 + "value": 1 } }, "8a3f3c76906944af8e4653d4cdc8e22e": { @@ -3105,13 +3027,13 @@ "description": "", "description_allow_html": false, "layout": "IPY_MODEL_ccfa2a1578f54a16b315cf45b553478c", - "max": 1.0, - "min": 0.0, + "max": 1, + "min": 0, "orientation": "horizontal", "style": "IPY_MODEL_45f4d8cd02c74b73a35d15e31fc33078", "tabbable": null, "tooltip": null, - "value": 1.0 + "value": 1 } }, "90b1a10420424e418ff34bf3dd4412cf": { diff --git a/setup.cfg b/setup.cfg index 6d6101d..91739e9 100644 --- a/setup.cfg +++ b/setup.cfg @@ -52,11 +52,12 @@ install_requires = click more_itertools wandb + huggingface_hub numpy torch pyfaidx genomepy - grelu>=1.0.10,<=1.0.11 + grelu>=1.0.10 lightning torchmetrics bioframe @@ -116,15 +117,16 @@ console_scripts = addopts = --cov decima --cov-report term-missing --verbose + -m "not wandb and not hf" norecursedirs = dist build .tox testpaths = tests # Use pytest markers to select/deselect specific tests -# markers = -# slow: mark tests as slow (deselect with '-m "not slow"') -# system: mark end-to-end system tests +markers = + wandb: mark tests that require wandb authentication (skipped by default) + hf: mark tests that call the HuggingFace API [devpi:upload] # Options for the devpi: PyPI server and packaging tool diff --git a/src/decima/constants.py b/src/decima/constants.py index af70f35..d885a35 100644 --- a/src/decima/constants.py +++ b/src/decima/constants.py @@ -3,6 +3,10 @@ import json import os +HF_MODEL_REPO = "Genentech/decima-model" +HF_DATA_REPO = "Genentech/decima-data" +HF_METADATA_FILENAME = "metadata.h5ad" + # constants for all models DECIMA_CONTEXT_SIZE = 524_288 diff --git a/src/decima/hub/__init__.py b/src/decima/hub/__init__.py index e50c657..de9aabe 100644 --- a/src/decima/hub/__init__.py +++ b/src/decima/hub/__init__.py @@ -1,32 +1,26 @@ -import os -import json +from pathlib import Path from typing import Union, Optional, List import warnings -import wandb -from pathlib import Path -from tempfile import TemporaryDirectory import anndata -from grelu.resources import get_artifact, DEFAULT_WANDB_HOST -from decima.constants import DEFAULT_ENSEMBLE, ENSEMBLE_MODELS, MODEL_METADATA +from huggingface_hub import hf_hub_download +from decima.constants import ( + DEFAULT_ENSEMBLE, + ENSEMBLE_MODELS, + MODEL_METADATA, + HF_MODEL_REPO, + HF_DATA_REPO, + HF_METADATA_FILENAME, +) from decima.model.lightning import LightningModel, EnsembleLightningModel -def login_wandb(): - """Login to wandb either as anonymous or as a user.""" - host = os.environ.get("WANDB_HOST", DEFAULT_WANDB_HOST) - try: - wandb.login(host=host, anonymous="never", timeout=0) - except wandb.errors.UsageError: # login anonymously if not logged in already - wandb.login(host=host, relogin=True, anonymous="must", timeout=0) - - def load_decima_model(model: Union[str, int, List[str]] = DEFAULT_ENSEMBLE, device: Optional[str] = None): - """Load a pre-trained Decima model from wandb or local path. + """Load a pre-trained Decima model from HuggingFace or a local path. Args: model: Model identifier or path. Can be: - int: Replicate number (0-3) - - str: Model name on wandb + - str: Model name from MODEL_METADATA - str: Path to local model checkpoint - List: list of local model checkpoints device: Device to load the model on. If None, automatically selects the best available device. @@ -40,56 +34,46 @@ def load_decima_model(model: Union[str, int, List[str]] = DEFAULT_ENSEMBLE, devi if isinstance(model, LightningModel): return model - elif model in ENSEMBLE_MODELS: + if model in ENSEMBLE_MODELS: return EnsembleLightningModel( [load_decima_model(model_name, device) for model_name in MODEL_METADATA[model]], name=model, ) - elif isinstance(model, List): + if isinstance(model, list): if len(model) == 1: return load_decima_model(model[0], device) - else: - return EnsembleLightningModel([load_decima_model(path, device) for path in model], name=model) + return EnsembleLightningModel([load_decima_model(path, device) for path in model], name=model) + + if isinstance(model, str) and Path(model).exists(): + if model.endswith(".ckpt"): + return LightningModel.load_from_checkpoint(model, map_location=device) + return LightningModel.load_safetensor(model, device=device) - # Load directly from a path if model in MODEL_METADATA: - model_name = MODEL_METADATA[model]["name"] - if "model_path" in MODEL_METADATA[model]: # if model path exist in metadata load it from the path + if "model_path" in MODEL_METADATA[model] and Path(MODEL_METADATA[model]["model_path"]).exists(): return load_decima_model(MODEL_METADATA[model]["model_path"], device) - elif isinstance(model, str) and Path(model).exists(): - if model.endswith("ckpt"): - return LightningModel.load_from_checkpoint(model, map_location=device) - else: - return LightningModel.load_safetensor(model, device=device) - else: - raise ValueError( - f"Invalid model: {model} it needs to be either a string of model_names on wandb (" - f"{list(MODEL_METADATA.keys())}), path to a local model, or a list of paths." - ) - # load model from wandb - art = get_artifact( - model_name, - project="decima", - host=os.environ.get("WANDB_HOST", DEFAULT_WANDB_HOST), + name = MODEL_METADATA[model]["name"] + cached = hf_hub_download(repo_id=HF_MODEL_REPO, filename=f"{name}.safetensors") + return LightningModel.load_safetensor(cached, device=device) + + raise ValueError( + f"Invalid model: {model}. Must be a known model name {list(MODEL_METADATA.keys())}, " + "a local path, or a list of local paths." ) - with TemporaryDirectory() as d: - art.download(d) - return LightningModel.load_safetensor(Path(d) / f"{model_name}.safetensors", device=device) def load_decima_metadata(name_or_path: Optional[str] = None): - """Load the Decima metadata from wandb. + """Load the Decima metadata from HuggingFace or a local path. Args: - name_or_path: Path to local metadata file or name of the model to load metadata for using wandb. If None, default model's metadata will be downloaded from wandb. + name_or_path: Path to local metadata file or model name. If None, downloads from HuggingFace. Returns: An AnnData object containing the Decima metadata. """ - if name_or_path is not None: - if Path(name_or_path).exists(): - return anndata.read_h5ad(name_or_path) + if name_or_path is not None and Path(name_or_path).exists(): + return anndata.read_h5ad(name_or_path) name_or_path = name_or_path or DEFAULT_ENSEMBLE @@ -98,22 +82,17 @@ def load_decima_metadata(name_or_path: Optional[str] = None): if name_or_path in MODEL_METADATA: metadata = MODEL_METADATA[name_or_path] - - if "metadata_path" in metadata: - if Path(metadata["metadata_path"]).exists(): - return anndata.read_h5ad(metadata["metadata_path"]) - else: + if "metadata_path" in metadata: + if Path(metadata["metadata_path"]).exists(): + return anndata.read_h5ad(metadata["metadata_path"]) warnings.warn( - f"Metadata `{metadata['metadata_path']}` provided in environment variables, " - f"but not found in `{metadata['metadata_path']}` " - f"Trying to download `metadata` from wandb." + f"Metadata path `{metadata['metadata_path']}` not found. Downloading from HuggingFace." ) + else: + raise ValueError( + f"Invalid metadata: {name_or_path}. Must be a known model name {list(MODEL_METADATA.keys())} " + "or a local path." + ) - art = get_artifact( - metadata["metadata"], - project="decima", - host=os.environ.get("WANDB_HOST", DEFAULT_WANDB_HOST), - ) - with TemporaryDirectory() as d: - art.download(d) - return anndata.read_h5ad(Path(d) / f"{metadata['metadata']}.h5ad") + cached = hf_hub_download(repo_id=HF_DATA_REPO, filename=HF_METADATA_FILENAME, repo_type="dataset") + return anndata.read_h5ad(cached) diff --git a/src/decima/hub/download.py b/src/decima/hub/download.py index 6788e74..e3d9a54 100644 --- a/src/decima/hub/download.py +++ b/src/decima/hub/download.py @@ -1,12 +1,17 @@ -import os +import logging from pathlib import Path from typing import Union -import logging import genomepy -from grelu.resources import get_artifact, DEFAULT_WANDB_HOST -from decima.constants import DEFAULT_ENSEMBLE, ENSEMBLE_MODELS, MODEL_METADATA -from decima.hub import login_wandb, load_decima_model, load_decima_metadata - +from huggingface_hub import hf_hub_download +from decima.constants import ( + DEFAULT_ENSEMBLE, + ENSEMBLE_MODELS, + MODEL_METADATA, + HF_MODEL_REPO, + HF_DATA_REPO, + HF_METADATA_FILENAME, +) +from decima.hub import load_decima_model, load_decima_metadata logger = logging.getLogger("decima") @@ -18,86 +23,75 @@ def cache_hg38(): def cache_decima_weights(): - """Download pre-trained Decima model weights from wandb.""" + """Download pre-trained Decima model weights from HuggingFace.""" logger.info("Downloading Decima model weights...") for rep in MODEL_METADATA[DEFAULT_ENSEMBLE]: load_decima_model(rep) def cache_decima_metadata(): - """Download pre-trained Decima model data from wandb.""" + """Download pre-trained Decima metadata from HuggingFace.""" logger.info("Downloading Decima metadata...") load_decima_metadata() def cache_decima_data(): """Download all required data for Decima.""" - login_wandb() cache_hg38() cache_decima_weights() cache_decima_metadata() def download_decima_weights(model: Union[str, int] = DEFAULT_ENSEMBLE, download_dir: str = "."): - """Download pre-trained Decima model weights from wandb. + """Download pre-trained Decima model weights from HuggingFace to a local directory. Args: - model_name: Model name or replicate number. - download_dir: Directory to download the model weights. + model: Model name or replicate number. + download_dir: Directory to save the model weights. Returns: Path to the downloaded model weights. """ if model in ENSEMBLE_MODELS: - return [download_decima_weights(model, download_dir) for model in MODEL_METADATA[model]] + return [download_decima_weights(m, download_dir) for m in MODEL_METADATA[model]] - model_name = MODEL_METADATA[model]["name"] + name = MODEL_METADATA[model]["name"] download_dir = Path(download_dir) download_dir.mkdir(parents=True, exist_ok=True) - logger.info(f"Downloading Decima model weights for {model} to {download_dir / f'{model_name}.safetensors'}") - - art = get_artifact(model_name, project="decima", host=os.environ.get("WANDB_HOST", DEFAULT_WANDB_HOST)) - art.download(str(download_dir)) - return download_dir / f"{model_name}.safetensors" + logger.info(f"Downloading Decima model weights for {model} to {download_dir / f'{name}.safetensors'}") + return Path(hf_hub_download(repo_id=HF_MODEL_REPO, filename=f"{name}.safetensors", local_dir=str(download_dir))) -def download_decima_metadata(metadata: str = DEFAULT_ENSEMBLE, download_dir: str = "."): - """Download pre-trained Decima model data from wandb. +def download_decima_metadata(download_dir: str = "."): + """Download pre-trained Decima metadata from HuggingFace to a local directory. Args: - download_dir: Directory to download the metadata. - metadata: Name of the model to download metadata for using wandb. + download_dir: Directory to save the metadata. Returns: Path to the downloaded metadata. """ - metadata = metadata or DEFAULT_ENSEMBLE - if metadata in ENSEMBLE_MODELS: - metadata = MODEL_METADATA[metadata][0] - - metadata_name = MODEL_METADATA[metadata]["metadata"] - art = get_artifact(metadata_name, project="decima", host=os.environ.get("WANDB_HOST", DEFAULT_WANDB_HOST)) download_dir = Path(download_dir) download_dir.mkdir(parents=True, exist_ok=True) - logger.info(f"Downloading Decima metadata to {download_dir / f'{metadata_name}.h5ad'}.") - - art.download(str(download_dir)) - return download_dir / f"{metadata_name}.h5ad" + logger.info(f"Downloading Decima metadata to {download_dir / HF_METADATA_FILENAME}") + return Path(hf_hub_download( + repo_id=HF_DATA_REPO, filename=HF_METADATA_FILENAME, repo_type="dataset", local_dir=str(download_dir) + )) def download_decima(model: str = DEFAULT_ENSEMBLE, download_dir: str = "."): """Download all required data for Decima. Args: - download_dir: Directory to download the model weights and metadata. + model: Model name or replicate number. + download_dir: Directory to save the model weights and metadata. Returns: - Path to the downloaded directory containing the model weights and metadata. + Path to the download directory. """ download_dir = Path(download_dir) download_dir.mkdir(parents=True, exist_ok=True) logger.info(f"Downloading Decima model weights and metadata to {download_dir}:") - download_decima_weights(model, download_dir) - download_decima_metadata(model, download_dir) + download_decima_metadata(download_dir) return download_dir diff --git a/src/decima/hub/wandb.py b/src/decima/hub/wandb.py new file mode 100644 index 0000000..9c0196a --- /dev/null +++ b/src/decima/hub/wandb.py @@ -0,0 +1,142 @@ +"""Wandb-based model loading for internal/private model access.""" + +import os +from pathlib import Path +from tempfile import TemporaryDirectory +from typing import Union, Optional, List +import warnings +import wandb +import anndata +from decima.constants import DEFAULT_ENSEMBLE, ENSEMBLE_MODELS, MODEL_METADATA +from decima.model.lightning import LightningModel, EnsembleLightningModel + +WANDB_ENTITY = "grelu" +WANDB_PROJECT = "decima" +DEFAULT_WANDB_HOST = "https://wandb.ai" + + +def _get_host(host: Optional[str] = None) -> str: + return host or os.environ.get("WANDB_HOST", DEFAULT_WANDB_HOST) + + +def login_wandb(host: Optional[str] = None): + """Login to wandb either as anonymous or as a user.""" + h = _get_host(host) + try: + wandb.login(host=h, anonymous="never", timeout=0) + except wandb.errors.UsageError: + wandb.login(host=h, relogin=True, anonymous="must", timeout=0) + + +def _get_artifact(name: str, host: Optional[str] = None): + api = wandb.Api(overrides={"host": _get_host(host)}) + return api.artifact(f"{WANDB_ENTITY}/{WANDB_PROJECT}/{name}:latest") + + +def load_decima_model( + model: Union[str, int, List[str]] = DEFAULT_ENSEMBLE, + device: Optional[str] = None, + host: Optional[str] = None, +): + """Load a pre-trained Decima model from wandb or a local path.""" + if isinstance(model, LightningModel): + return model + + if model in ENSEMBLE_MODELS: + return EnsembleLightningModel( + [load_decima_model(m, device, host) for m in MODEL_METADATA[model]], + name=model, + ) + + if isinstance(model, list): + if len(model) == 1: + return load_decima_model(model[0], device, host) + return EnsembleLightningModel([load_decima_model(p, device, host) for p in model], name=model) + + if isinstance(model, str) and Path(model).exists(): + if model.endswith(".ckpt"): + return LightningModel.load_from_checkpoint(model, map_location=device) + return LightningModel.load_safetensor(model, device=device) + + if model in MODEL_METADATA: + if "model_path" in MODEL_METADATA[model]: + return load_decima_model(MODEL_METADATA[model]["model_path"], device, host) + name = MODEL_METADATA[model]["name"] + art = _get_artifact(name, host) + with TemporaryDirectory() as d: + art.download(d) + return LightningModel.load_safetensor(Path(d) / f"{name}.safetensors", device=device) + + raise ValueError( + f"Invalid model: {model}. Must be a known model name {list(MODEL_METADATA.keys())}, " + "a local path, or a list of local paths." + ) + + +def load_decima_metadata(name_or_path: Optional[str] = None, host: Optional[str] = None): + """Load the Decima metadata from wandb or a local path.""" + if name_or_path is not None and Path(name_or_path).exists(): + return anndata.read_h5ad(name_or_path) + + name_or_path = name_or_path or DEFAULT_ENSEMBLE + + if name_or_path in ENSEMBLE_MODELS: + name_or_path = MODEL_METADATA[name_or_path][0] + + if name_or_path in MODEL_METADATA: + metadata = MODEL_METADATA[name_or_path] + if "metadata_path" in metadata: + if Path(metadata["metadata_path"]).exists(): + return anndata.read_h5ad(metadata["metadata_path"]) + warnings.warn( + f"Metadata path `{metadata['metadata_path']}` not found. Downloading from wandb." + ) + art = _get_artifact(metadata["metadata"], host) + with TemporaryDirectory() as d: + art.download(d) + return anndata.read_h5ad(Path(d) / f"{metadata['metadata']}.h5ad") + + raise ValueError(f"Unknown model: {name_or_path}") + + +def download_decima_weights( + model: Union[str, int] = DEFAULT_ENSEMBLE, + download_dir: str = ".", + host: Optional[str] = None, +): + """Download pre-trained Decima model weights from wandb.""" + if model in ENSEMBLE_MODELS: + return [download_decima_weights(m, download_dir, host) for m in MODEL_METADATA[model]] + + name = MODEL_METADATA[model]["name"] + download_dir = Path(download_dir) + download_dir.mkdir(parents=True, exist_ok=True) + art = _get_artifact(name, host) + art.download(str(download_dir)) + return download_dir / f"{name}.safetensors" + + +def download_decima_metadata( + metadata: str = DEFAULT_ENSEMBLE, + download_dir: str = ".", + host: Optional[str] = None, +): + """Download pre-trained Decima metadata from wandb.""" + metadata = metadata or DEFAULT_ENSEMBLE + if metadata in ENSEMBLE_MODELS: + metadata = MODEL_METADATA[metadata][0] + + metadata_name = MODEL_METADATA[metadata]["metadata"] + download_dir = Path(download_dir) + download_dir.mkdir(parents=True, exist_ok=True) + art = _get_artifact(metadata_name, host) + art.download(str(download_dir)) + return download_dir / f"{metadata_name}.h5ad" + + +def cache_decima_data(host: Optional[str] = None): + """Download all required Decima assets from wandb.""" + login_wandb(host) + for rep in MODEL_METADATA[DEFAULT_ENSEMBLE]: + load_decima_model(rep, host=host) + load_decima_metadata(host=host) diff --git a/tests/conftest.py b/tests/conftest.py index 82d5336..97ad5e5 100644 --- a/tests/conftest.py +++ b/tests/conftest.py @@ -6,7 +6,6 @@ import pytest from decima.constants import DECIMA_CONTEXT_SIZE -from decima.hub import login_wandb from decima.hub.download import cache_hg38 @@ -38,7 +37,6 @@ def pytest_collection_modifyitems(config, items): item.add_marker(skip_long_running) -login_wandb() cache_hg38() diff --git a/tests/test_hub.py b/tests/test_hub.py index 0bcda93..0063c6c 100644 --- a/tests/test_hub.py +++ b/tests/test_hub.py @@ -2,9 +2,16 @@ import anndata from decima.model.lightning import LightningModel from decima.hub import load_decima_model, load_decima_metadata +from decima.constants import HF_MODEL_REPO, HF_DATA_REPO, HF_METADATA_FILENAME -@pytest.mark.long_running +def test_hf_constants(): + assert HF_MODEL_REPO == "Genentech/decima-model" + assert HF_DATA_REPO == "Genentech/decima-data" + assert HF_METADATA_FILENAME == "metadata.h5ad" + + +@pytest.mark.hf def test_load_decima_model(): model_0 = load_decima_model() assert model_0 is not None @@ -14,6 +21,7 @@ def test_load_decima_model(): assert model_2 is not None +@pytest.mark.hf def test_load_decima_metadata(): metadata = load_decima_metadata() assert isinstance(metadata, anndata.AnnData)