omicverse.single.MetaCell

omicverse.single.MetaCell#

class omicverse.single.MetaCell(adata, method='seacells', use_rep='X_pca', n_metacells=None, layer=None, device='cpu', random_state=0, **kwargs)[source]#

Unified metacell wrapper with dispatchable backends.

Parameters:

  • adata – Input single-cell AnnData.

  • method (str (default: 'seacells')) – Backend key. One of 'seacells' (default, backward-compatible), 'metaq', 'mc2', 'supercell', 'kmeans', 'random', 'geosketch'.

  • use_rep (str (default: 'X_pca')) – Embedding key in adata.obsm. Used by graph-based backends (seacells/supercell/kmeans/geosketch). MetaQ and MC2 derive their own representations and ignore this.

  • n_metacells (Optional[int] (default: None)) – Target number of metacells. Default adata.n_obs // 75.

  • layer (Optional[str] (default: None)) – Counts layer for backends that need raw counts (MetaQ, MC2).

  • device (str (default: 'cpu')) – 'cpu', 'cuda', or 'mps' for GPU-capable backends.

  • random_state (int (default: 0)) – Seed forwarded to all backends.

  • **kwargs – Forwarded to the backend constructor. See each backend’s docstring for valid kwargs.

__init__(adata, method='seacells', use_rep='X_pca', n_metacells=None, layer=None, device='cpu', random_state=0, **kwargs)[source]#
Parameters:

  • method (str (default: 'seacells'))

  • use_rep (str (default: 'X_pca'))

  • n_metacells (Optional[int] (default: None))

  • layer (Optional[str] (default: None))

  • device (str (default: 'cpu'))

  • random_state (int (default: 0))

Methods

__init__(adata[, method, use_rep, ...])

assign_new_cells(adata_query)

capability_matrix()

check_rigor([layer_lognorm, feature_use, ...])

Score the rigor of the current partition (mcRigor port).

codebook()

compactness([use_rep])

Legacy alias → compute_compactness.

compute_celltype_purity([celltype_label])

Legacy alias → compute_purity.

compute_compactness([use_rep])

Per-metacell mean pairwise distance in use_rep (lower = more compact).

compute_purity([label_key])

Per-metacell majority-label purity.

compute_separation([use_rep, label_key])

Mean intra/inter-metacell distance ratio per metacell (lower = more separated).

fit(**kwargs)

Train the chosen backend; write unified schema into adata.

fit_multi_gamma(gammas)

initialize_archetypes(**kwargs)

Legacy SEACells shim.

latent()

load(path)

predicted([method, layer, summary, ...])

Aggregate single cells into a metacell-level AnnData.

save(path)

separation([use_rep, nth_nbr])

Legacy alias → compute_separation.

soft_membership()

step([n_steps])

Legacy SEACells shim — only meaningful for the seacells backend.

train([min_iter, max_iter])

Legacy SEACells shim → .fit().

Attributes

capabilities

Contents