Deconvolving cell-type-specific gene expression profiles from bulk RNA-seq samples

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Bulk RNA sequencing (RNA-seq) provides average gene expression at low cost.
• Single-cell RNA sequencing (scRNA-seq) offers high-resolution transcriptomics but at a higher cost.
• Integrating bulk RNA-seq data with scRNA-seq insights is crucial for comprehensive analysis.

Purpose of the Study:

• To develop a deep learning algorithm, BLUE, for deconvolving bulk RNA-seq samples.
• To accurately predict cell-type proportions and cell-type-specific gene expression profiles.
• To leverage these predictions for cancer patient subtyping and biomarker discovery.

Main Methods:

• Developed a U-Net-based deep learning algorithm named BLUE.
• Utilized BLUE's feature extraction for accurate transcriptomic deconvolution.
• Applied the algorithm to predict cell-type-specific gene expression from bulk RNA-seq data.

Main Results:

• BLUE significantly outperforms existing deconvolution algorithms in predicting cell-type-specific gene expression.
• The algorithm accurately estimates cell-type proportions from bulk RNA-seq data.
• Achieved accurate predictions enabling downstream applications in cancer research.

Conclusions:

• BLUE effectively integrates bulk RNA-seq data strengths with single-cell resolution insights.
• The developed framework facilitates cancer patient subtyping using deconvolution results.
• Identified cell-type-specific gene signatures as potential prognostic biomarkers for cancer.