A Bayesian Noisy Logic Model for Inference of Transcription Factor Activity from Single Cell and Bulk Transcriptomic Data

Abstract

The advent of high-throughput sequencing has made it possible to measure the expression of genes at relatively low cost. However, direct measurement of regulatory mechanisms, such as Transcription Factor (TF) activity is still not readily feasible in a high-throughput manner. Consequently, there is a need for computational approaches that can reliably estimate regulator activity from observable gene expression data. In this work, we present a noisy Boolean logic Bayesian model for TF activity inference from differential gene expression data and causal graphs. Our approach provides a flexible framework to incorporate biologically motivated TF-gene regulation logic models. Using simulations and controlled over-expression experiments in cell cultures, we demonstrate that our method can accurately identify TF activity. Moreover, we apply our method to bulk and single cell transcriptomics measurements to investigate transcriptional regulation of fibroblast phenotypic plasticity. Finally, to facilitate usage, we provide user-friendly software packages and a web-interface to query TF activity from user input differential gene expression data: https://umbibio.math.umb.edu/nlbayes/. Author SummaryNextGen RNA sequencing (RNA-Seq) has enabled simultaneous measurement of the expression level of all genes. Measurements can be done at the population level or single-cell resolution. However, direct measurement of regulatory mechanisms, such as Transcription Factor (TF) activity, is still not possible in a high-throughput manner. As such, there is a need for computational models to infer regulator activity from gene expression data. In this work, we introduce a Bayesian methodology that utilizes prior biological knowledge on bio-molecular interactions in conjunction with readily available gene expression measurements to estimate TF activity. The Bayesian model naturally incorporates biologically motivated combinatorial TF-gene interaction logic models and accounts for noise in gene expression data as well as prior knowledge. The method is accompanied by efficiently implemented R and Python software packages as well as a user-friendly web-based interface that allows users to upload their gene expression data and run queries on a TF-gene interaction network to identify and rank putative transcriptional regulators. This tool can be used for a wide range of applications, such as identification of TFs downstream of signaling events and environmental or molecular perturbations, the aberration in TF activity in diseases, and other studies with case-control gene expression data.