Estimating Heterogeneous Gene Regulatory Networks from Zero-Inflated Single-Cell Expression Data

• Activity Number: 55 - Statistical methods for data from single cell technologies
• Type: Contributed
• Date/Time: Sunday, August 8, 2021 : 3:30 PM to 5:20 PM
• Sponsor: Section on Statistics in Genomics and Genetics
• Author(s): Qiuyu Wu* and Xiangyu Luo
• Companies: Renmin University of China and Renmin University of China
• Keywords: Bayesian analysis; Gaussian graphical models; heterogeneity; nonignorable dropout; spike-slab prior

Abstract

Inferring gene regulatory networks can elucidate how genes work cooperatively. The gene-gene collaboration information is often learned by Gaussian graphical models (GGM) that aim to identify whether the expression levels of any pair of genes are dependent given other genes' expression values. One basic assumption that guarantees the validity of GGM is data normality, and this often holds for bulk-level expression data which aggregate biological signals from a collection of cells. However, fine-grained cell-level expression profiles collected in single-cell RNA-sequencing (scRNA-seq) reveal non-normality features---cellular heterogeneity and zero-inflation. We propose a Bayesian latent mixture GGM to jointly estimate multiple gene regulatory networks accounting for the zero-inflation and unknown heterogeneity of single-cell expression data. The proposed approach outperforms competing methods on synthetic data in terms of network structure and precision matrix estimation accuracy and provides biological insights when applied to two real-world scRNA-seq datasets.