adataobsm

adataobsm refers to the obsm attribute in an AnnData object, a core data container used in the Scanpy ecosystem for single-cell transcriptomics. The obsm attribute stores observation-level multi-dimensional arrays that encode per-cell coordinates or embeddings produced by dimensionality reduction or embedding methods.

Data model: It behaves like a dictionary mapping strings to 2D arrays with shape (n_obs, k). Each value represents coordinates in a reduced space for the same set of observations. All arrays in obsm share the same number of rows (n_obs). The keys typically identify the embedding or coordinates, such as X_pca, X_umap, X_tsne, or X_diffmap, as well as any user-defined embeddings.

Common usage: Typical keys include X_pca, X_umap, X_tsne, and X_diffmap. The corresponding arrays contain the coordinates for each observation (cell) in the specified embedding space. Access is via adata.obsm['X_umap'] to obtain the UMAP coordinates (n_obs by 2). These coordinates are commonly used for visualization, clustering, or downstream analyses that rely on a consistent cell representation across methods.

Storage and interoperability: obsm arrays are stored when saving an AnnData to disk (for example in .h5ad format) and are preserved across sessions. They are typically dense NumPy arrays used for plotting, neighborhood graph construction, and other analyses. obsm is distinct from obsp, which holds pairwise relationships such as distances or adjacency graphs between observations.

Notes: many analysis steps populate obsm automatically (for example, umap, tsne, and pca computations in Scanpy), but users can also add or modify entries manually. The obsm container is part of the broader AnnData API, alongside obs (per-observation annotations), var (per-variable annotations), and uns (unstructured annotations).