advaitpaliwal scvi-tools
Use this skill for reproducible scvi-tools analysis.
Workflow:
- Record AnnData path, organism, assay, batch keys, label keys, covariates, train/test split, and filtering choices.
- Verify Python environment, GPU/CPU route, package version, and data availability before training.
- Save preprocessing notebook, model parameters, training logs, latent embeddings, differential-expression tables, and plots.
- Check batch mixing, biological separation, marker consistency, and sensitivity to preprocessing choices.
- Attach exact commands and artifact paths to the final summary.
Keep raw counts, normalized values, and model-derived latent variables clearly separated.