Single-Cell Differential Network Analysis with Sparse Bayesian Factor Models.

Michael Sekula,Susmita Datta,Jeremy Gaskins

doi:10.3389/fgene.2021.810816

Michael Sekula, Susmita Datta + Show 1 more

Open Access

https://doi.org/10.3389/fgene.2021.810816

Copy DOI

Abstract

Differential network analysis plays an important role in learning how gene interactions change under different biological conditions, and the high resolution of single-cell RNA (scRNA-seq) sequencing provides new opportunities to explore these changing gene-gene interactions. Here, we present a sparse hierarchical Bayesian factor model to identify differences across network structures from different biological conditions in scRNA-seq data. Our methodology utilizes latent factors to impact gene expression values for each cell to help account for zero-inflation, increased cell-to-cell variability, and overdispersion that are unique characteristics of scRNA-seq data. Condition-dependent parameters determine which latent factors are activated in a gene, which allows for not only the calculation of gene-gene co-expression within each group but also the calculation of the co-expression differences between groups. We highlight our methodology’s performance in detecting differential gene-gene associations across groups by analyzing simulated datasets and a SARS-CoV-2 case study dataset.

Highlights

Gene network modeling has become essential to the understanding of complex biological systems related to health and disease
We propose a hierarchical Bayesian factor model for constructing gene co-expression networks (GCNs) from scRNA-seq data to explore differences in the network structure across various cell groups due to different biological conditions, cell types, cell stages, or other group choice
The NORmal To Anything (NORTA) algorithm generates a random vector from a multivariate standard normal distribution with a given correlation structure and transforms it into a random vector with a specified marginal distribution

Summary

Introduction

Gene network modeling has become essential to the understanding of complex biological systems related to health and disease. These networks allow researchers to uncover and interpret relationships and interactions between genes during different biological processes (Blencowe et al, 2019). The vast majority of these methods have focused only on analyzing gene expressions from one cellular population, such as a single tissue type, disease, or environmental condition. In the context of bulk population data (i.e., microarray and bulk RNA sequencing), efforts have been made to develop different strategies for identifying differences between gene-gene networks. Some approaches propose qualitative analyses through visual inspection of different network topologies (Caldana et al, 2011; Weston et al, 2011), while others rely on statistical tests to determine differences across conditions (Choi and Kendziorski, 2009; Gill et al, 2010; Fukushima, 2013)

Methods

Results

Discussion

Conclusion