A novel Bayesian framework for harmonizing information across tissues and studies to increase cell type deconvolution...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate cell type deconvolution from bulk transcriptomics relies on robust reference signature matrices.
• Single-cell RNA sequencing (scRNA-seq) offers a rich resource for deriving these signatures, but is challenged by noise.
• Existing methods for signature matrix inference from scRNA-seq data can be limited.

Purpose of the Study:

• To introduce tranSig, a novel Bayesian framework for improved signature matrix inference from scRNA-seq data.
• To leverage shared cell type-specific expression patterns across diverse tissues and studies.
• To enhance the accuracy of computational cell type deconvolution in bulk transcriptomics.

Main Methods:

• Developed a novel Bayesian framework named tranSig.
• Utilized scRNA-seq data to infer cell type signature matrices.
• Leveraged shared cross-tissue and cross-study expression patterns.
• Validated through simulations and applications to real-world bulk RNA sequencing datasets.

Main Results:

• tranSig demonstrates robustness to variations in the number of signature genes and tissues.
• The framework accurately characterizes biological heterogeneity across sample groups.
• Applications to peripheral blood, bronchoalveolar lavage, and aorta data show significant improvements in deconvolution.

Conclusions:

• tranSig provides an accurate and robust method for defining cell type gene expression signatures.
• This facilitates more precise in silico cell type deconvolution from bulk transcriptomics.
• The approach enhances the analysis of biological heterogeneity in complex tissues.