MitoOmics-GPU [Work in Progress]

GPU-accelerated multi-omics pipeline to quantify and visualize the Mitochondrial Health Index (MHI) by integrating extracellular vesicle/mitochondrial-derived vesicle (EV/MDV) proteomics with single-cell RNA-seq.

Hackathon project by Team Go Getters at the NVIDIA Accelerate Omics Hackathon (8-25 Sept 2025).

👥 Team Go Getters

• Sayane Shome, PhD (AI in Healthcare, Stanford)[Team Lead]
• Seema Parte, PhD (Ophthalmology, Stanford)
• Hirenkumar Patel, PhD (Ophthalmology, Stanford)
• Ankit Maisuriya (PhD candidate, Quantum Photonics, Northeastern)
• Medha Bhattacharya (CS undergrad, UC Irvine)

---

🚀 Project Objective

• Develop a GPU-accelerated pipeline for mitochondrial health analysis.

• Link blood-derived EV/MDV proteomics with mitochondrial DNA copy-number proxies from scRNA-seq.

• Provide interpretable measures:

  • Biogenesis (capacity to grow new mitochondria)
  • Fusion/Fission (structural remodeling)
  • Mitophagy (repair/recycling)
  • Heterogeneity (variation across cells).

• Output: a unified Mitochondrial Health Index (MHI) summarizing mitochondrial resilience, fitness, and disease risk.

---

⚡ Installation

pip install mitoomics-gpu

---

🖥️ GPU Acceleration

• Optimized with RAPIDS + GPU backends.
• Clear CPU vs GPU speedups for large datasets.
• Open-source, designed for integration with scverse/rapids-singlecell.

📊 Key Insights

• Unified mitochondrial health scoring (MHI).
• Patient-level and cell-type–level insights.
• Supports biomarker discovery, disease progression prediction, and drug response stratification.

---

---

📂 Real Data (Bundled)

All datasets are pre-bundled under mitoomics_gpu/data/:

File	Description
data/scrna.h5ad	Full scRNA-seq dataset (AnnData, with subject_id, cell_type, batch)
data/scrna.mito.h5ad	Mito-filtered scRNA-seq (pre-subsetted to mitochondrial genes)
data/ev_human.csv	EV/MDV proteomics from PRIDE (PXD018301) — columns: subject_id, protein, abundance
data/mitocarta3_table.csv	MitoCarta 3.0 pathway table (parsed from Human.MitoCarta3.0.xls)
data/genesets_curated.csv	Curated gene sets (fusion, fission, mitophagy, biogenesis)
data/ev_whitelist.csv	EV-specific protein whitelist for filtering

---

🧬 Usage with Real Data

Standard CPU pipeline

python -m mitoomics_gpu \
  --scrna        mitoomics_gpu/data/scrna.h5ad \
  --proteomics   mitoomics_gpu/data/ev_human.csv \
  --mitocarta-table mitoomics_gpu/data/mitocarta3_table.csv \
  --ev-whitelist mitoomics_gpu/data/ev_whitelist.csv \
  --outdir       results/

GPU-accelerated pipeline (recommended for large datasets)

python -m mitoomics_gpu.gpu_cli \
  --scrna        mitoomics_gpu/data/scrna.h5ad \
  --proteomics   mitoomics_gpu/data/ev_human.csv \
  --genesets_csv mitoomics_gpu/data/genesets_curated.csv \
  --ev_whitelist mitoomics_gpu/data/ev_whitelist.csv \
  --outdir       results/ \
  --do_umap

Using the mito-filtered scRNA (faster, recommended)

Replace scrna.h5ad with scrna.mito.h5ad in either command above to use the pre-filtered mitochondrial gene subset, which significantly reduces memory and compute time:

python -m mitoomics_gpu.gpu_cli \
  --scrna        mitoomics_gpu/data/scrna.mito.h5ad \
  --proteomics   mitoomics_gpu/data/ev_human.csv \
  --genesets_csv mitoomics_gpu/data/genesets_curated.csv \
  --ev_whitelist mitoomics_gpu/data/ev_whitelist.csv \
  --outdir       results/

Outputs written to results/:

• results_summary.csv / results_summary_GPU.csv — subject-level MHI scores
• embedding_pca_GPU.csv — PCA embedding (GPU run)
• embedding_umap_GPU.csv — UMAP embedding (if --do_umap passed)
• report.md + figures — visual summary

🔮 Future Directions

• Add modalities: scATAC, metabolomics, spatial transcriptomics.
• Deploy web-server / pip package for biologist-friendly use.
• Clinical validation with partners & cohorts.
• ML upgrades for pattern discovery & prediction on MHI.

📬 Contact

📧 sshome@stanford.edu