installation.md

Create conda env

Create and activate conda environment

conda env create -f environment.yml
conda activate sc-blipper

Note: If you want to give the env a different name: conda env create -f environment.yml -n my_env

Note down the install path for the conda env, we will need it later for configuring the pipeline

echo $CONDA_PREFIX  # On Unix/Linux/macOS

Exit conda env

conda deactivate

I cannot use conda, now what?

You may be able to create a singularity/apptainer container based on the conda enviroment, As of writing I have not tested this. Some resources below:

• https://arcdocs.leeds.ac.uk/arc3-arc4/usage/conda-containers.html
• https://stackoverflow.com/questions/76146763/create-apptainer-container-with-environment-yml-without-creating-a-new-conda-env
• https://csc-training.github.io/csc-env-eff/hands-on/singularity/singularity_extra_replicating-conda.html

Once you have your containers, you can set the parameters

params.rn_container="/path/to/container"
params.scvi.container="/path/to/container-scvi"

Also make sure to tell nextflow to use your container engine of choice, example for singularity:

conda.enabled = false
singularity.enabled = true

More details: https://www.nextflow.io/docs/latest/container.html#singularity

Install scVI (optional)

We will make a new enviroment for scVI to avoid conflicts and keep scVI optional so the pipeline is lighter You will ideally need a GPU for scVI to run

conda env create -f environment_scvi.yml
conda activate sc-blipper-scvi 

Note down the install path for the conda env, we will need it later for configuring the pipeline

echo $CONDA_PREFIX  # On Unix/Linux/macOS

At this point I strongly recommend testing GPU is working. If on HPC, make sure to request a GPU node You can test by running python and typing:

import torch
print(torch.cuda.is_available()) # Should return True
import jax
print(jax.devices()) # Should show GPU devices

Exit conda env

conda deactivate

If this doesn't work

Remove the conda enviroment you just created

conda remove -n sc-blipper-scvi --all

Manually create a new one

conda create -n sc-blipper-scvi python==3.10
conda activate sc-blipper-scvi

Follow instructions here: https://docs.scvi-tools.org/en/1.0.0/installation.html Install Pytorch, JAX, they might need to be tweaked depending on your system and GPU Pytorch: https://pytorch.org/get-started/locally/ JAX: https://docs.jax.dev/en/latest/installation.html#installation

Cuda 12 on linux, make sure the CUDA version matches your GPU drivers

pip3 install torch torchvision
pip3 install -U "jax[cuda12]"
pip install scvi-tools scikit-misc ipython