Robust Personalized Gene Co-Expression Network Construction from Single Cell RNA-Seq Data

• Activity Number: 527 - Novel Statistical Methods for Single-Cell Genomic Data
• Type: Invited
• Date/Time: Thursday, August 11, 2022 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Genomics and Genetics
• Author(s): Sunduz Keles* and Shan Lu
• Companies: University of Wisconsin, Madison and University of Wisconsin, Madison
• Keywords: scRNA-seq; co-expression; network; measurement error; iPSC differentiation

Abstract

Gene co-expression networks have been critical in exploring the relationships between different genes, gene sets, or between gene sets and clinical features. While methods for constructing co-expression networks from bulk RNA-seq data are well established, accurately estimating such networks from single cell transcriptome data is challenging due to sparsity and heterogeneity of transcript counts. We developed pGCN (personalized gene co-expression network) to estimate personalized gene co-expression networks from scRNA-seq data of individual subjects. We show that gene-gene correlations estimated from normalized transcript counts are severely biased and that co-expression networks are highly sensitive to varying numbers of cells and sequencing depths. pGCN mitigates these challenges by developing a measurement error model to compute gene-gene correlations and de-biasing network module traits. Application of pGCN on an scRNA study of long-term human induced pluripotent stem cell (iPSC) differentiation across multiple donors captures network level variability of donors that correlate with differentiation efficiency.