eSVD-DE: cohort-wide differential expression in single-cell RNA-seq data using exponential-family embeddings

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA) data is increasingly used in clinical and cohort studies.
• Existing methods struggle to identify differentially expressed (DE) genes in large scRNA datasets due to individual variation.
• Analyzing large cohorts presents challenges in accounting for individual-level confounding covariates and sparsely observed genes.

Purpose of the Study:

• To develop a novel method for differential gene expression analysis in large single-cell RNA sequencing cohorts.
• To address the challenges posed by individual variation and confounding covariates in cohort-wide DE analysis.
• To improve the accuracy and power of DE gene detection in large-scale scRNA studies.

Main Methods:

• Developed eSVD-DE, a matrix factorization approach to pool information across genes and remove confounding covariates.
• Implemented a novel two-sample test for mean expression differences between case and control groups.
• Utilized a hierarchical model to test differences in posterior mean distributions, mitigating Type-1 error inflation.

Main Results:

• eSVD-DE demonstrated higher accuracy and power compared to existing methods repurposed for cohort-wide DE analysis.
• The method effectively pools information across genes and accounts for individual-level confounding effects.
• Successfully overcomes Type-1 error inflation often seen in dimension reduction-based differential testing.

Conclusions:

• eSVD-DE offers a powerful new approach for testing differential gene expression in large scRNA cohorts post-dimension reduction.
• Accurate individual-level differential expression identification is vital for connecting scRNA-seq findings to human population understanding.
• This method enhances the utility of scRNA sequencing data in large-scale clinical and population studies.