muudata

MuData is a data container designed for multi-omics single-cell data. It provides a unified structure for storing several related AnnData objects that represent different molecular modalities for the same set of cells, enabling integrated analyses while preserving modality-specific information.

Structure and data model: A MuData object wraps multiple modality-specific AnnData objects (for example RNA, ATAC, and protein). Each modality has its own data matrix X, feature annotations var, and cell annotations obs, while a shared set of observations (cells) links the modalities. The container can also store per-modality uns metadata and cross-modality mappings to facilitate integration workflows. MuData structures are typically serialized to the .h5mu file format, which supports lazy loading and efficient I/O.

Interoperability and usage: MuData is used within the Python ecosystem alongside AnnData and Scanpy. It enables loading, subsetting, and processing of complex multi-omics datasets in a single object, supporting joint analyses such as integrated dimension reduction, clustering, and multi-view integration while maintaining modality-specific data.

Applications and context: The MuData container was developed to address the needs of multi-omics single-cell studies, where researchers collect complementary modalities (for example transcriptomics and chromatin accessibility) from the same or related cells. It is part of the broader AnnData ecosystem and is commonly employed in workflows that require coherent management of multiple modalities and their metadata.

Limitations and notes: MuData assumes a coherent mapping of cells across modalities or a clear correspondence strategy. For very large datasets, memory usage and I/O considerations are important. Documentation and example workflows are available in the MuData repository and related tutorials.