anndata

Annotated data matrices can be handled in backed mode to process large single-cell datasets without loading everything into memory, enabling efficient loading, filtering, and transformation across X, obs, and var components.

Using the AnnData structure allows you to store the main data matrix X alongside embeddings in obsm, unstructured metadata in uns, and feature annotations in var, keeping all single-cell analysis components aligned in one object.

Yes, the framework supports broad I/O compatibility for reading and writing h5ad, zarr, loom, and CSV-like formats, ensuring annotated data matrices integrate smoothly across different single-cell analysis environments.

Yes, the framework is designed for integration with the Scanpy ecosystem, allowing you to pass annotated data matrices directly into downstream single-cell analysis workflows and tools like Muon or PyTorch.

You can store processing history and version information in the uns attribute, and use the raw component to preserve pre-transformation data matrices, ensuring full traceability and reproducibility during single-cell analysis.

Backed mode requires compatible file formats like h5ad or zarr to access data on disk, and while it enables memory-efficient handling of large datasets, certain in-memory transformations or modifications might be restricted or require loading data fully.