scvi-tools

1# scvi-tools — scVI / scANVI
2 
3scvi-tools (Gayoso et al. 2022, github.com/scverse/scvi-tools, BSD-3-Clause)
4wraps a family
5of deep generative models for single-cell omics. The scRNA-seq core is **scVI**
6(unsupervised batch-corrected latent embedding) and **scANVI** (scVI + a
7classifier head for semi-supervised cell-type label transfer). Both expect
8**raw integer UMI counts** and emit a low-dimensional `X_scVI` / `X_scANVI`
9that drops into the scanpy neighbors → leiden → umap pipeline.
10 
11## Setup (any agent, no API key)
12 
13This is a **pure skill** — `kernel.py` is deterministic Python and _you_ (the
14base model) do all the reasoning. There is no `host` runtime and no LLM API.
15The helpers are `prepare_scvi_counts` for input validation and `h5ad_safe_obs`
16for obs/var frames that `anndata.write_h5ad()` can serialize. Load them once
17per session in a Python cell:
18 
19```python
20exec(open("scvi-tools/kernel.py", encoding="utf-8").read())   # path to this skill's kernel.py
21```
22 
23Nothing auto-loads it outside Claude Science. Then call the helpers directly.
24If a helper raises `NameError`, you haven't exec'd kernel.py.
25 
26Dependencies: `pip install scvi-tools scanpy anndata`. Training needs a
27CUDA-capable GPU — see [Remote compute](#remote-compute-rent-a-gpu) to fall out
28to a rented GPU when you don't have one locally.
29 
30## How to run
31 
32### scVI — batch-corrected latent space
33 
34`prepare_scvi_counts(adata)` checks and preserves existing `counts`. If absent,
35…