scvi-tools
by K-Dense-AIscvi-tools is a Python framework for probabilistic single-cell analysis. Use this scvi-tools skill for batch correction, latent embeddings, differential expression with uncertainty, transfer learning, and multimodal integration. It is a strong fit for single-cell RNA-seq, ATAC, CITE-seq, multiome, and spatial workflows, especially for advanced Machine Learning use cases.
This skill scores 78/100, which means it is a solid listing candidate for directory users: it is clearly triggerable, covers real single-cell analysis workflows, and gives enough operational framing to justify installation, though it still has some gaps in executable guidance and supporting assets.
- Strong trigger clarity for single-cell use cases: batch correction, multimodal integration, differential expression, transfer learning, and spatial transcriptomics are explicitly named.
- Substantial workflow content: the SKILL.md body is large, structured, and includes multiple headings plus code fences, which suggests more than a placeholder.
- Good install decision value: the description clearly positions scvi-tools versus generic analysis tools like scanpy, helping users know when this skill is the right fit.
- No install command, scripts, or support files are provided, so agents may still need to infer setup or execution details.
- The repository appears to be documentation-heavy without external references/resources, which limits trust signals and makes deeper validation harder.
Overview of scvi-tools skill
What scvi-tools is for
The scvi-tools skill helps you use scVI-style probabilistic models for single-cell omics when a normal analysis prompt is too vague. It is most useful for batch correction, latent representation learning, integration across runs or donors, and uncertainty-aware differential expression. If your goal is advanced single-cell modeling rather than routine preprocessing, this scvi-tools skill is a strong fit.
Who should install it
Install scvi-tools if you work with single-cell RNA-seq, multiome, CITE-seq, ATAC, or spatial data and want a model-driven workflow. It is especially relevant for Machine Learning users who need a PyTorch-based framework, not just a static method summary. If you mainly need basic QC, clustering, or visualization, a standard Scanpy-first workflow is usually enough.
What matters before adoption
The main value is not just that scvi-tools exists, but that it gives you a practical path from raw counts to trained latent models with explicit tradeoffs. The key decision is whether you need probabilistic modeling, transfer learning, or multimodal integration enough to justify the added setup and modeling choices. This skill is worth installing when output quality depends on handling batch effects or comparing heterogeneous datasets carefully.
How to Use scvi-tools skill
Install the skill
Use the directory install flow for the scvi-tools skill:
npx skills add K-Dense-AI/claude-scientific-skills --skill scvi-tools
After install, verify the skill path under scientific-skills/scvi-tools and open the source file directly. For this repository, SKILL.md is the primary entry point; there are no supporting rules/, resources/, or scripts/ folders to rely on.
Read the right files first
Start with SKILL.md to understand scope, model families, and the recommended decision points. Then scan the sections on when to use the skill, core capabilities, and the single-cell RNA-seq workflow before attempting a prompt. Because the repository is compact, the fastest way to reduce guesswork is to read the file end to end once rather than cherry-pick only the model names.
Turn a rough goal into a usable prompt
A weak request like “analyze my scRNA-seq data” is not enough. A better scvi-tools usage prompt names the assay, data shape, and decision you need:
- “Use scVI to integrate 6 scRNA-seq batches, compare donor effects, and return the latent space plus batch-mixing diagnostics.”
- “Fit a MULTIVI-style workflow for paired RNA + ATAC data and explain whether cells are better separated by biology or batch.”
- “Run differential expression with uncertainty on two cell populations and report effect sizes, not just p-values.”
Workflow tips that change output quality
Give the skill the input it needs to choose the right model family: modality, number of batches, whether data are paired, and whether the task is integration, annotation, or DE. State any constraints up front, such as sparse counts, small sample size, or the need to stay compatible with existing scanpy objects. When you want the best scvi-tools guide result, ask for the model choice, setup steps, expected outputs, and common failure modes in one pass.
scvi-tools skill FAQ
Is scvi-tools only for scRNA-seq?
No. The scvi-tools skill covers multiple single-cell modalities, including RNA-seq, ATAC, multimodal assays, and spatial use cases. That said, RNA integration is the most common entry point, so it is usually the easiest place to validate fit before expanding to more complex data.
Do I need this if I already use Scanpy?
Use both, but for different jobs. Scanpy is better for standard preprocessing and exploratory workflows, while scvi-tools is better when you need probabilistic modeling, latent embeddings, or integration under batch effects. If your analysis question does not require a learned generative model, scvi-tools may be more than you need.
Is this beginner-friendly?
It is beginner-accessible only if you already understand basic single-cell concepts like counts matrices, batches, and annotations. The skill helps most when you can specify your data and goal clearly. If you cannot yet say whether you need integration, transfer learning, or differential expression, start with a simpler analysis path first.
When should I not use scvi-tools?
Do not reach for scvi-tools for simple normalization, quick plotting, or one-off exploratory checks. It is also a poor fit if you want a purely statistical cookbook without model selection decisions. For tiny datasets or highly custom pipelines, the overhead may outweigh the benefit.
How to Improve scvi-tools skill
Give the model-selection context
The biggest quality gain comes from telling the skill what kind of scvi-tools problem you actually have. Say whether you need scVI, TOTALVI, MultiVI, or another family only after describing the data, not before. For example, “paired CITE-seq with strong donor effects” is more useful than “use MultiVI.”
Share data structure and constraints
Better inputs reduce the most common failure mode: choosing the wrong model for the assay. Include matrix type, number of cells, batches, covariates, and whether counts are raw or normalized. If you are working in a scvi-tools for Machine Learning workflow, also mention whether you want a reusable latent space, downstream classifier features, or an interpretable comparison against another model.
Ask for outputs you can act on
Do not ask only for “analysis.” Ask for a concrete deliverable such as a training plan, a model choice rationale, integration diagnostics, or a notebook-style workflow. If the first result is too generic, iterate by adding what was missing: cell type labels, batch definitions, or what you need to compare against scanpy or another baseline.