omicverse.bulk.Deconvolution

omicverse.bulk.Deconvolution(adata_bulk, adata_single, max_single_cells: int = 5000, celltype_key: str = 'celltype', cellstate_key: str = None, gpu: int | str = 0)[source]

Bulk RNA-seq deconvolution class for inferring cell-type fractions from single-cell references.

Parameters:

  • adata_bulk (AnnData) – Bulk expression matrix with samples in rows and genes in columns.

  • adata_single (AnnData) – Single-cell reference matrix containing cell-level expression profiles and cell-type annotations used to build signature profiles.

  • max_single_cells (int) – Maximum number of cells to keep from adata_single. If the reference contains more cells, a random subset is used to control memory/runtime.

  • celltype_key (str) – Column name in adata_single.obs storing cell-type labels.

  • cellstate_key (str or None) – Optional column name in adata_single.obs storing finer cell-state labels (used by methods such as BayesPrism).

  • gpu (Union[int,str]) – Compute device selector. Supports CUDA index (for example 0), explicit strings such as 'cuda:0', 'mps', or 'cpu'.

Returns:

Initializes deconvolution inputs and builds reference expression profiles.

Return type:

None

Examples

>>> deconv_obj = ov.bulk.Deconvolution(adata_bulk, adata_single, celltype_key="celltype")