Towards causality in gene regulatory network inference

Understanding the coordination of biomolecules that underlies gene regulation is key to gaining mechanistic insights into cellular functions, phenotypes, and diseases. Advances in single-cell technologies promise to unveil mechanisms of gene regulation at unprecedented resolution by enabling measure...

Full description

Bibliographic Details
Main Author: Wu, Alexander Po-Yen
Other Authors: Berger, Bonnie A.
Format: Thesis
Published: Massachusetts Institute of Technology 2023
Online Access:https://hdl.handle.net/1721.1/151203
Description
Summary:Understanding the coordination of biomolecules that underlies gene regulation is key to gaining mechanistic insights into cellular functions, phenotypes, and diseases. Advances in single-cell technologies promise to unveil mechanisms of gene regulation at unprecedented resolution by enabling measurements of genomic and/or epigenetic features for individual cells. However, unlocking insights from single-cell data requires algorithmic innovations. This thesis introduces a series of methods for uncovering gene regulatory relationships underlying cellular identity and function from single-cell data. Firstly, we present a framework for enhancing the detection of statistical associations in small sample size settings for gene regulatory network inference. We then describe the use of single-cell genetic perturbation screens for determining the causal roles of critical regulatory complexes, focusing specifically on its applications for revealing mechanistic insights about the mammalian SWI/SNF family of chromatin remodeling complexes. To bridge the gap between methods that identify statistical associations from observational data and those that infer causal relationships using interventions, we also introduce a new category of techniques that extends the econometric concept of Granger causality to complex graph-based dynamical systems, such as those found in single-cell trajectories. In particular, we describe a graph neural network-based generalization of Granger causality for single-cell multimodal data that enables the detection of noncoding genomic loci implicated in the regulation of specific genes. We then demonstrate how we use this approach to link genetic variants to gene dysregulation in disease, focusing on its applications to schizophrenia etiology. Lastly, we present an extension of this graph-based Granger causal framework that leverages RNA velocity dynamics for causal gene regulatory network inference and enables inquiries into the role of temporal control in gene regulatory function and disease.