omicverse.bulk2single.Bulk2Single

omicverse.bulk2single.Bulk2Single#

class omicverse.bulk2single.Bulk2Single(bulk_data, single_data, celltype_key, bulk_group=None, max_single_cells=5000, top_marker_num=500, ratio_num=1, gpu=0)[source]#

VAE-based bulk-to-single framework for reconstructing pseudo single cells from bulk RNA-seq.

Parameters:

  • bulk_data (pd.DataFrame) – Bulk expression matrix with genes in rows and samples in columns.

  • single_data (anndata.AnnData) – Reference single-cell dataset used to learn cell-type expression patterns.

  • celltype_key (str) – Column in single_data.obs containing cell-type labels.

  • bulk_group (Optional[Any]) – Optional sample grouping information for averaging bulk replicates.

  • max_single_cells (int) – Maximum number of reference cells retained for model fitting.

  • top_marker_num (int) – Number of marker genes per cell type used by downstream preparation.

  • ratio_num (int) – Multiplier controlling total generated cell counts.

  • gpu (Union[int,str]) – Device selector for training (CUDA index, 'mps', or CPU fallback).

Returns:

Initializes bulk2single deconvolution and simulation workflow.

Return type:

None

Examples

>>> model = ov.bulk2single.Bulk2Single(bulk_data=bulk_data, single_data=single_data, celltype_key="Cell_type")
__init__(bulk_data, single_data, celltype_key, bulk_group=None, max_single_cells=5000, top_marker_num=500, ratio_num=1, gpu=0)[source]#

Initialize the Bulk2Single class for bulk-to-single-cell deconvolution.

Parameters:

  • bulk_data (pd.DataFrame) – Bulk RNA-seq expression matrix. Rows are genes and columns are samples.

  • single_data (anndata.AnnData) – Single-cell reference with compatible gene symbols and cell metadata.

  • celltype_key (str) – Name of the column in single_data.obs that stores cell-type labels.

  • bulk_group (Optional[Any]) – Optional grouping key/list used to aggregate replicate bulk samples.

  • max_single_cells (int) – Maximum number of reference cells used during initialization.

  • top_marker_num (int) – Intended number of marker genes per cell type for preprocessing steps.

  • ratio_num (int) – Generation ratio used when estimating target cell counts.

  • gpu (Union[int,str]) – Compute device specification; supports CUDA indices and 'mps'.

Methods

__init__(bulk_data, single_data, celltype_key)

Initialize the Bulk2Single class for bulk-to-single-cell deconvolution.

bulk_preprocess_lazy()

Preprocess bulk RNA-seq data for deconvolution.

filtered(generate_adata[, ...])

Filter generated single-cell data by removing low-quality clusters.

generate()

Generate synthetic single-cell data from trained VAE model.

load(vae_load_dir[, hidden_size])

Load a pre-trained VAE model.

load_and_generate(vae_load_dir[, hidden_size])

Load pre-trained VAE model and generate single-cell data.

load_fraction(fraction_path)

Load predicted cell-type target numbers from file.

plot_loss([figsize])

Plot training loss curve of the VAE model.

predicted_fraction([method, sep, scaler, ...])

Predict cell-type fractions from bulk RNA-seq data using deconvolution.

prepare_input()

Prepare input data for VAE training.

save([vae_save_dir, vae_save_name])

Save the trained VAE model and cell target numbers.

single_preprocess_lazy([target_sum])

Preprocess single-cell reference data.

train([vae_save_dir, vae_save_name, ...])

Train the VAE model for single-cell data generation.
Contents