A starter stack for single-cell analysis

Single-cell data is deceptively simple — a big matrix of cells by genes — and surprisingly subtle to analyze well. Here is a dependable stack for getting from raw counts to biological insight.

Pick an ecosystem

You have two excellent options, and most labs standardize on one:

• Python / scverse: AnnData as the data structure, Scanpy for the standard workflow, and scvi-tools for probabilistic deep-learning models.
• R / Bioconductor: Seurat for an all-in-one toolkit, backed by the broader Bioconductor ecosystem.

A typical Scanpy workflow

1import scanpy as sc
2adata = sc.read_h5ad("pbmc.h5ad")
3sc.pp.filter_cells(adata, min_genes=200)
4sc.pp.normalize_total(adata, target_sum=1e4)
5sc.pp.log1p(adata)
6sc.pp.highly_variable_genes(adata, n_top_genes=2000)
7sc.pp.pca(adata)
8sc.pp.neighbors(adata)
9sc.tl.leiden(adata)
10sc.pl.umap(adata, color="leiden")
Copy

When to reach for deep learning

Batch effects and multi-dataset integration are where probabilistic models earn their keep. scVI learns a shared latent space across batches; scANVI adds semi-supervised label transfer. These integrate cleanly with the AnnData objects Scanpy produces.

Don't collect data you could download

Before you sequence anything, check CZ CELLxGENE Census — tens of millions of harmonized cells you can query and pull straight into AnnData. Reference atlases make annotation faster and your conclusions stronger.