Key Takeaways
- scRNA-seq analysis requires 8+ distinct steps — each benefiting from specialized AI skills
- K-Dense scRNA-seq skills provide the most comprehensive coverage: QC, normalization, clustering, DE, trajectory
- MedgeClaw bundles 140 K-Dense skills including full scRNA-seq pipeline support
- Skills encode expert decisions: why SCTransform over log-normalization, optimal Leiden resolution, appropriate statistical tests
- Key tools wrapped: Scanpy, Seurat, scVI, CellRanger, Harmony, scVelo
The scRNA-seq Analysis Pipeline
| Step | What It Does | Key Decision | Skill Coverage |
|---|---|---|---|
| 1. QC | Filter low-quality cells/genes | Threshold selection (nGenes, %mito) | K-Dense ✓ |
| 2. Normalization | Scale expression values | SCTransform vs log-normalization | K-Dense ✓ |
| 3. HVG Selection | Identify variable genes | Number of HVGs, method | K-Dense ✓ |
| 4. Batch Correction | Remove technical variation | Harmony vs BBKNN vs scVI | K-Dense ✓ |
| 5. Dimensionality Reduction | PCA → UMAP/t-SNE | Number of PCs | K-Dense ✓ |
| 6. Clustering | Group similar cells | Resolution parameter | K-Dense ✓ |
| 7. Differential Expression | Find marker genes | Statistical test selection | K-Dense ✓ |
| 8. Trajectory Inference | Order cells in pseudotime | Root cell selection | K-Dense ✓ |
| 9. Cell Type Annotation | Label clusters | Reference database choice | K-Dense ✓ |
Skill Libraries for scRNA-seq
| Library | scRNA-seq Skills | Tool Coverage | Explainability |
|---|---|---|---|
| K-Dense | 20+ | Scanpy, Seurat, scVI, Harmony, scVelo | High (parameter justification) |
| MedgeClaw bundle | 140 (including scRNA) | Full bioinformatics suite | High (research dashboard) |
| Orchestra | 3-5 (general) | General data analysis | Low (ML-focused) |
FAQ
Q1: Should I use Scanpy or Seurat skills?
K-Dense provides skills for both. Scanpy skills are more comprehensive due to Python ecosystem integration. Use Seurat skills if your lab standardizes on R.
Q2: How do skills handle batch correction?
The batch correction skill evaluates your data structure (number of batches, batch size, confounding) and recommends the appropriate method. It explains why Harmony was chosen over BBKNN for your specific dataset.
Q3: Can skills handle spatial transcriptomics too?
Yes. K-Dense includes separate spatial transcriptomics skills for spot deconvolution, spatial pattern analysis, and spatially variable gene detection.
Summary
scRNA-seq analysis involves 8+ steps, each requiring expert decisions. K-Dense scRNA-seq skills provide the most comprehensive coverage — encoding expert knowledge about parameter selection, method choice, and result validation. For a complete setup, use MedgeClaw (bundles 140 skills) or install K-Dense skills individually into your preferred agent.