Debiased personalized gene coexpression networks for population-scale scRNA-seq data

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Population-scale single-cell RNA sequencing (scRNA-seq) offers insights into gene expression variation.
• Estimating gene coexpression networks from scRNA-seq is challenging due to technical noise and sparse, low-expression data.
• Existing methods struggle with zero-biased correlation estimates for genes with low expression.

Purpose of the Study:

• To present Dozer, a novel method for debiasing gene-gene correlation estimates from scRNA-seq data.
• To enable accurate quantification of network-level gene expression variation across individuals.
• To improve the accuracy and reliability of gene coexpression networks derived from scRNA-seq.

Main Methods:

• Dozer corrects correlation estimates using a general Poisson measurement model.
• It provides a metric to identify genes with high measurement noise.
• The method was evaluated through computational experiments and applied to population-scale scRNA-seq datasets.

Main Results:

• Dozer produces robust correlation estimates, unaffected by mean expression levels or sequencing depth.
• It significantly reduces false-positive edges in coexpression networks compared to alternatives.
• Dozer improves the accuracy of network centrality measures, module detection, and batch integration.

Conclusions:

• Dozer is a significant advancement for estimating personalized coexpression networks from scRNA-seq data.
• The method enables novel analyses, such as identifying gene groups linked to induced pluripotent stem cell differentiation efficiency.
• Dozer revealed distinct coexpression modules in Alzheimer's disease and control brain tissue, highlighting immune response differences.