omicverse-bulk-celltype-deconvolution

To deconvolve bulk RNA-seq, provide an AnnData-based bulk cohort and a paired single-cell atlas to infer cell-type fractions using algorithms like TAPE, Scaden, or BayesPrism. The interface uses cell type and optional cell state keys to generate per-sample fraction tables and grouped visualizations.

You can compare TAPE, Scaden, and BayesPrism by running them through a unified deconvolution interface on the same bulk cohort and single-cell reference. This allows direct evaluation of resulting cell fraction estimates and validation of inferred proportions against canonical categories.

Yes, BayesPrism supports hierarchical cell-state modeling for bulk RNA-seq deconvolution. Provide a single-cell atlas with coarse cell types and optional cell states to infer both broad cell-type fractions and detailed cell-state proportions simultaneously.

Yes, bulk RNA-seq deconvolution works directly with AnnData-based bulk cohorts and single-cell atlas references. The interface requires AnnData inputs to map bulk expression profiles against single-cell data, producing per-sample cell fraction estimates and grouped composition visualizations.

Estimate immune cell proportions from a bulk PBMC cohort by using a paired single-cell PBMC reference to deconvolve the bulk RNA-seq data. The workflow infers immune cell fractions, compares Bayesian and deep-learning backends, and validates inferred proportions across phenotype-level samples.

TAPE and Scaden bulk deconvolution parameters include n_cores for parallel processing, fast_mode for accelerated computation, and pseudobulk_size for reference aggregation. Specify the celltype_key and optional cellstate_key in the single-cell reference to generate accurate cell fractions.