CGMFinder Identifies Correlated Gene Modules from 3H scRNA-seq Data

Abstract

Correlated gene modules (CGMs) contain genes whose expression fluctuates together. Genes in CGMs are often functionally related and regulated by shared transcription factors. CGMs can be identified under steady-state conditions in populations of cells using single-cell RNA sequencing (scRNA-seq). Here, we introduce CGMFinder, a tool for CGM identification using "3H" scRNA-seq data (High mRNA capture efficiency, High cell numbers, and High sequencing depth). CGMFinder employs a graph-based filtering approach, first identifying CGM cores from highly-expressed genes and then linking noisy low-abundance genes to these cores. In lymphoblastoid cell line 3H datasets generated by in-lab and commercial protocols, CGMFinder accurately identifies CGMs enriched for gene ontologies or pathways. In cells grown under hypoxic conditions, CGMFinder successfully identified hypoxia-specific "glycolysis" and "response to oxygen levels" modules. Evaluations using ground truth correlation modules demonstrate that CGMFinder outperforms other CGM identification methods such as WGCNA and FastICA in scRNA-seq data.