Transcriptional regulatory genomics: from mechanistic modeling to causal inference

Gene transcription refers to the process in which coding regions on the genome are copied into mRNA molecules through complex cellular mechanisms. This process is regulated through mechanisms that are encoded in the genome and are activated by cellular signals, enzymes, and proteins. Transcriptional regulation often involves a class of proteins called transcription factors regulating other genes. Although we have a limited understanding of regulatory mechanisms and associations in human and other species, discerning such characteristics of transcriptional regulation is of paramount importance in systems biology. The recent advancements in experimental techniques for high throughput measurements of cellular processes and their molecular signatures have led to increasingly growing biological databases including various ``omics" datasets. These resources give rise to the emergence of novel computational models in systems biology that aim at understanding the genome of human and other species from data. These efforts include the development of data-driven methods for modeling transcriptional regulation using omics datasets. The general goal of such studies is to understand regulatory mechanisms and molecular interactions that drive transcriptional regulation. In practice, both the predictive accuracy and interpretability of these quantitative models are crucial to improve their efficacy. Especially, interpretability of the model is a key factor in various applications, from learning mechanistic regulatory insights to inferring causal regulatory relationships. This thesis is focused on the applications of interpretable computational models in learning and simulating transcriptional regulatory systems. In this Ph.D. thesis, I develop novel interpretable machine learning models for studying transcriptional regulations from two aspects: (1) learning biophysically-consistent regulatory mechanisms, (2) inference of causal regulatory associations. The first aspect of the study was pursued through quantitative and machine learning models that either explicitly encode regulatory mechanisms using biophysically-inspired functions or learn them in meaningful higher-order representations. The second aspect was achieved through an interpretation of non-linear machine learning models based on causal inference principles. Additionally, I leverage an existing mechanistic model for stochastic expression of genes to develop a novel framework for simulating gene expressions under causal regulatory networks at the cell-level resolution. This tool is useful for assessing the strength and weaknesses of causal regulatory inference algorithms.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2023

Author dc:creator

• Dibaeinia, Payam

Contributors dc:contributor

• Sinha, Saurabh
• Zhai, ChengXiang
• El-Kebir, Mohammed
• Dresch, Jacqueline M.

• Gene Regulation
• Gene Regulatory Networks
• Machine Learning
• Interpretable Ai For Biology
• Causal Inference
• Counterfactual Inference

Statement dc:rights

• Copyright 2023 Payam Dibaeinia

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/120101

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F120101