Cellxgene Census

This gives you direct access to 61+ million standardized single-cell genomics records from the CZ CELLxGENE Census. You can filter by cell type, tissue, or disease, then pull expression matrices into scanpy for analysis or pipe them into PyTorch dataloaders for training ML models. The API handles both in-memory queries for smaller datasets and streaming iterators when you're working with millions of cells that won't fit in RAM. It's built on top of SOMA, so you get versioned snapshots for reproducibility. If you're doing population-scale single-cell work or need large labeled datasets for cell type classification, this beats scraping individual studies from GEO.