Primary Submission Category: Applications in Health and Biology
Inferring Directed Gene Regulatory Networks from Single-Cell Perturb-Seq Data and Poisson-Log Normal Models
Authors: Zhongxuan Sun, Hyunseung Kang, Sunduz Keles,
Presenting Author: Zhongxuan Sun*
We present a new framework for inferring directed gene regulatory networks (GRNs) with a novel Poisson-log normal (PLN) model for single-cell Perturb-Seq data, a new type of experiment in genomics where for each cell (i.e., the study unit), CRISPR-based technology is used to either silence a particular gene (i.e., the treated cell) or not (i.e., the control cell). Unlike existing methods for discovery of GRNs that rely on acyclicity and normality assumptions, our PLN-based framework accommodates count-based, overdispersed transcriptomic data and potential feedback loops, an important feature of single-cell Perturb-Seq data. Additionally, unlike most existing methods for GRN discovery which rely on observational expressional data, the estimates from the PLN-based framework are not biased from unmeasured confounders due to the design of Perturb-Seq experiments and accurately capture regulatory directionality. We validate our inferred networks through complementary epigenomic and proteomic evidence: ChIP-seq datasets confirm the presence of regulatory interactions at key promoter and enhancer regions, and protein–protein interaction (PPI) networks that carry functional relationships among regulators and their targets. These multi-modal validations underscore the robustness of our method for reconstructing large-scale GRNs. Our results highlight the power of single-cell perturbation data, integrated with comprehensive molecular profiling, to reveal intricate regulatory circuitry