Bulk RNA-seq deconvolution with Bulk2Single

Bulk RNA-seq deconvolution extracts single-cell resolution information from aggregate data by estimating cell-type fractions and generating synthetic single-cell profiles using a beta-VAE model trained against a reference scRNA-seq atlas.

Yes, this Skill uses integrated TAPE estimators to accurately estimate cell-type fractions directly from bulk RNA-seq samples, providing quantified proportions of different cell types present in your tissue.

After generating synthetic cells, the Skill benchmarks them by comparing cluster proportions and expression correlations against your reference scRNA-seq data, enabling quality assessment of deconvolved results.

You need omicverse, scanpy, scvelo, anndata, matplotlib, numpy, pandas, and torch installed. GPU support is available but optional, with explicit control over resource usage during model training and synthetic cell generation.

Yes, the Skill handles bulk tumor RNA-seq by deconvolving immune and tumor cell-type composition into synthetic single-cell profiles, enabling detailed cell-type proportion analysis and validation against known atlases.

The Skill harmonizes bulk and reference scRNA-seq atlas data, aligns feature spaces, performs lazy preprocessing to manage large datasets, and handles error recovery with resumption capabilities for robust pipeline execution.