muon

muon extends the AnnData model to multimodal experiments — CITE-seq, multiome, and anything else measuring more than one modality in the same cells. It exists because single-cell work stopped being about transcriptomics alone, and a data structure built for one matrix stops fitting when there are three.

What it does

A MuData object holds several AnnData objects side by side, keeps their observations aligned, and supports both per-modality and joint analysis, including multi-omics factor analysis and joint embeddings. Modalities can be processed with their own appropriate methods and then integrated, rather than forced into a single pipeline.

Strengths

• Native multimodal container that keeps modalities properly aligned.
• Per-modality processing with joint downstream integration.
• Interoperates with Scanpy and scvi-tools.
• On-disk H5MU format for multimodal interchange.

Limitations

It adds a layer of abstraction on top of an already layered structure, and the learning curve is real. The multimodal ecosystem is less mature than the RNA-only one, so methods and best practice are still moving. If your experiment is single-modality, this is unnecessary complexity — stay with plain AnnData.