Skills scvi-tools

📦

scvi-tools

Name: scvi-tools
Author: K-Dense-AI

Low Risk 🌐 Network access📁 Filesystem access

Analyze single-cell genomics data

Also available from: davila7

Single-cell genomics analysis requires probabilistic models to handle technical noise and batch effects. scvi-tools provides deep generative models using variational inference for dimensionality reduction, batch correction, differential expression, and multi-modal integration of single-cell data.

Supports: Claude Codex Code(CC)

📊 70 Adequate

Download the skill ZIP

Upload in Claude

Go to Settings → Capabilities → Skills → Upload skill

Toggle on and start using

Test it

Using "scvi-tools". Help me set up scVI for batch correction on my single-cell dataset with two batches.

Expected outcome:

```python
import scvi
import scanpy as sc

# Preprocess data
adata = sc.read_h5ad("my_single_cell_data.h5ad")
sc.pp.filter_genes(adata, min_counts=3)

# Register data with batch information
scvi.model.SCVI.setup_anndata(
adata,
layer="counts",
batch_key="batch"
)

# Train model
model = scvi.model.SCVI(adata)
model.train()

# Get batch-corrected latent representation
latent = model.get_latent_representation()
adata.obsm["X_scVI"] = latent

# Downstream analysis
sc.pp.neighbors(adata, use_rep="X_scVI")
sc.tl.umap(adata)
sc.tl.leiden(adata)
```

Using "scvi-tools". How do I identify marker genes between two cell types in my scVI model?

Expected outcome:

```python
# Differential expression between two groups
de_results = model.differential_expression(
groupby="leiden",
group1="0", # Cluster 0
group2="1", # Cluster 1
mode="change",
delta=0.25 # Minimum effect size
)

# View top differentially expressed genes
print(de_results.head(20))

# Filter for significant genes
significant_genes = de_results[
(de_results['is_de_fdr_0.05']) &
(de_results['bayes_factor'] > 1)
]
print(f"Found {len(significant_genes)} differentially expressed genes")
```

Security Audit

Low Risk

v5 • 1/21/2026

This is a documentation-only skill containing markdown reference files for scvi-tools, a legitimate Python library for single-cell genomics analysis. All 399 static findings are false positives caused by incorrect pattern matching: Python code examples in documentation were flagged as shell commands, bioinformatics statistical terms were misidentified as cryptographic algorithms, and documentation URLs were flagged as hardcoded URLs. No executable code or malicious patterns exist. Safe for publication.

Files scanned

7,698

Lines analyzed

findings

Total audits

Risk Factors

🌐 Network access (1)

SKILL.md:167-169

📁 Filesystem access

No specific locations recorded

Audited by: claude View Audit History →

Quality Score

Architecture

100

Maintainability

Content

Community

Security

Spec Compliance

What You Can Build

Batch correction for integrated single-cell analysis

Remove technical batch effects from single-cell RNA-seq datasets across multiple donors, protocols, or sequencing runs using scVI to create unified, integrated cell atlases.

Differential expression with uncertainty

Identify differentially expressed genes between cell types or conditions with probabilistic uncertainty estimates, providing more reliable statistical conclusions for downstream validation.

Multi-modal data integration

Jointly analyze paired RNA and protein measurements (CITE-seq) or chromatin accessibility data to discover cell populations with enhanced biological resolution.

Try These Prompts

Basic scVI model setup

Help me set up scvi-tools to analyze my single-cell RNA-seq data. I have an AnnData object with raw count data and want to perform batch correction. Show me how to register the data, train the model, and extract latent representations.

Differential expression analysis

I trained an scVI model on my single-cell dataset with cell type annotations. Help me identify differentially expressed genes between two cell types (e.g., cluster A vs cluster B) using the differential_expression method. Include how to interpret the results and set effect size thresholds.

Multi-modal integration with totalVI

I have paired CITE-seq data with RNA counts and protein antibody-derived counts. Help me set up totalVI to jointly model both modalities, train the model, and extract joint latent representations that capture both RNA and protein variation.

Spatial transcriptomics deconvolution

I have a single-cell reference dataset with cell type annotations and a spatial transcriptomics dataset with spot-level counts. Help me use DestVI or Stereoscope to deconvolve cell types in the spatial data and create cell type proportion maps.

Best Practices

Always provide raw, unnormalized count data to scvi-tools models for accurate probabilistic modeling
Register all known technical covariates (batch, donor, protocol) during setup to improve batch correction
Save trained models regularly using model.save() to avoid retraining on large datasets
Use GPU acceleration (accelerator="gpu") when training on datasets with more than 50,000 cells

Avoid

Do not use log-normalized data as input - scvi-tools models expect raw count data
Do not skip data filtering (low-count genes/cells) before training as it affects model quality
Do not interpret latent representations without validation against known biological markers
Do not use scvi-tools for bulk RNA-seq analysis - it is designed specifically for single-cell data

Frequently Asked Questions

What is the difference between scVI and scanpy for single-cell analysis?

scvi-tools uses deep generative models with variational inference to learn probabilistic representations of single-cell data, providing uncertainty quantification and superior batch correction. scanpy provides traditional dimensionality reduction (PCA, UMAP) and clustering methods. Use scvi-tools for advanced modeling and batch correction, scanpy for standard pipelines and downstream analysis.

Do I need a GPU to use scvi-tools?

A GPU is recommended but not required. For datasets with fewer than 50,000 cells, CPU training is feasible. For larger datasets or frequent analysis, GPU acceleration significantly reduces training time. Install with GPU support using: pip install scvi-tools[cuda]

What data formats does scvi-tools support?

scvi-tools works with AnnData objects from the scanpy ecosystem. Data should be stored in the .X matrix as raw counts. Cell-level covariates (batch, donor) and gene-level information are registered using model.setup_anndata(). Support for Seurat objects is available through conversion.

How do I choose between scVI, scANVI, and totalVI?

Use scVI for unsupervised analysis without cell type labels. Use scANVI when you have some cell type annotations and want semi-supervised integration. Use totalVI for paired RNA and protein (CITE-seq) data to jointly model both modalities.

What does the differential_expression method return?

It returns a DataFrame with gene-level statistics including: bayes_factor (evidence for differential expression), is_de_fdr_0.05 (FDR-corrected significance), mean (average expression per group), and effect size estimates. Higher bayes_factor indicates stronger evidence for differential expression.

How do I save and load trained scvi-tools models?

Use model.save("path/to/model", overwrite=True) to save and model = scvi.model.SCVI.load("path/to/model", adata=new_adata) to load. The saved model includes model weights, training history, and configuration. This allows sharing models and avoiding retraining.