univi.workflows.stack_embeddings

univi.workflows.stack_embeddings(model, datasets, *, device='cpu', batch_size=1024)

Encode (cohort, modality, AnnData) entries and stack modality means.

Each input cell contributes one row per supplied modality. Features in the result are latent dimensions, not a union of incompatible assay features. cell_id retains the input barcode and block the tuple’s cohort label. No UMAP is fit and no biological pairing is inferred by this function.