Rigorous probabilistic methods for analyzing diverse omics data to provide insights into cell states and dynamics

Ghaffari, Saba

Permalink

https://hdl.handle.net/2142/129905 Copy

Description

Title

Rigorous probabilistic methods for analyzing diverse omics data to provide insights into cell states and dynamics

Author(s)

Ghaffari, Saba

Issue Date

2025-06-18

Director of Research (if dissertation) or Advisor (if thesis)

Sinha, Saurabh

Doctoral Committee Chair(s)

Sinha, Saurabh

Committee Member(s)

• Zhai, Cheng-Xiang
• El-Kebir, Mohammed
• Offer, Steven M

Department of Study

Computer Science

Discipline

Computer Science

Degree Granting Institution

University of Illinois Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

multi-omics integrated analysis, probabilistic graphical models, Bayesian probabilistic modelling, cancer metastasis, transcriptional regulation, gene regulatory network, cellular dynamics, single cell transcriptomics, bulk gene expression deconvolution, dynamic epigenome

Abstract

Our cells consist of a complex network of molecules that interact with each other in different ways to maintain normal cellular processes. Gene regulation is the cornerstone of cellular processes that controls when, where, and to what extent genes are expressed in different cellular states. Disruption of the molecular interactions that regulate gene expression can lead to altered cell states, as observed in the development or progression of different diseases such as cancer. Identifying perturbed molecular interactions and their affected direct or indirect targets is crucial as it can provide insight into novel therapeutic design. This necessitates a holistic view of the underlying molecular interactions driving the cellular regulatory system. Therefore, in the past decade, much effort has been focused on large-scale profiling of different types of molecules, termed multi-omics, under varying cellular conditions. Additionally, powerful computational approaches are required to discover patterns of disrupted interactions and their driving mechanisms, by aggregating information across multiple types of molecular measurements. This thesis proposes a comprehensive interpretable probabilistic approach for the integration of multi-omics data, measured for bulk population of cells, in different cellular states. The proposed approach can identify gene regulatory elements whose altered function is associated with cellular state dynamics. Its application to colorectal cancer progression successfully identified a transcription factor whose knockdown was experimentally validated to reduce both cancer progression and invasion in colorectal cancer cell line. A small set of genes predicted by the model formed a strong signature to differentiate cohorts of colorectal cancer patients with different survival outcomes, demonstrating future opportunities for the model findings to be used in clinical translation. The accuracy of modeling regulatory dynamics and their alteration relies heavily on the granularity of the cellular measurements. Single cell technologies advanced over the past decade allow for cellular level measurements of different omics. However, bulk measurements, aggregated over population of cells, are more frequently adopted under limited budget. This demands the development of computational techniques that can infer cell type-specific expression profiles and cell type composition of samples derived from bulk transcriptomics. This thesis proposes a comprehensive Bayesian probabilistic approach to deconvolve bulk expression profiles given reference cell type signatures, into shared adjusted cell type specific expression profiles and sample specific cell type proportions. The new approach is shown to be especially robust to noise in the reference signatures, a common challenge for current deconvolution methods. Application of the proposed method to a rare genetic disorder called dihydropyridine dehydrogenase deficiency identified the possible involvement of ciliopathy and impaired translational control in its etiology. In summary, this thesis shows how to incorporate knowledge of gene regulation into the design of interpretable powerful probabilistic models that can accurately identify perturbed molecular interactions as part of the regulatory system underlying altered cell states.

Graduation Semester

2025-08

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/129905

Copyright and License Information

Copyright 2025 Saba Ghaffari

Owning Collections