patient-data-ops

Patient Data Operations

This skill ensures consistent handling of synthetic patient data and bioinformatics datasets.

Data Hierarchy

Always use these standard paths:

• data/patient-data/: Individual patient .h5ad or .csv files (e.g., PAT001-OVC-2025/)
• data/reports/: Generated analysis reports (PDF/Markdown)
• data/images/: Visualizations (UMAP, Heatmaps, PCA)

Handling scRNA-seq Data (.h5ad)

The perturbation server (mcp-perturbation) works with AnnData objects for GEARS predictions. When working in that context:

1. Normalization: Ensure scanpy.pp.normalize_total and scanpy.pp.log1p are applied
2. HVG Selection: Standardize on 7,000 Highly Variable Genes (HVG) for GEARS compatibility
3. Metadata: Preserve cell_type and condition keys in .obs

Data Preprocessing Tools

Use these standard scripts where possible:

• generate_synthetic_data.py: For creating test-safe patient mocks
• data_loader.py: Use the loaders in src/mcp_[name] to ensure consistent I/O

Data Privacy & Security

• Anonymization: Never store PII (Patient Identifiable Information) in the repo
• Mocking: Always use synthetic data for unit tests
• Synthetic only: This repo contains only synthetic patient data (HIPAA safe)

Data Quality Checklist

Before running simulations:

• Check for NaN values in expression matrices
• Verify cell-type labels match the project ontology
• Ensure the file size fits within the 4Gi Cloud Run limit

---

Use this skill when:

• Loading new patient datasets
• Preprocessing scRNA-seq files for GEARS (perturbation server context)
• Creating test data for bioinformatics tools