Machine learning and data analytics for liver disease modeling

Hu, Chang

Permalink

https://hdl.handle.net/2142/132636 Copy

Description

Title

Machine learning and data analytics for liver disease modeling

Author(s)

Hu, Chang

Issue Date

2025-11-05

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

Iyer, Ravishankar K

Doctoral Committee Chair(s)

Iyer, Ravishankar K

Committee Member(s)

• Anastasio, Mark A
• Milenkovic, Olgica
• Shomorony, Ilan
• Lazaridis, Konstantinos N
• Wang, Liewei

Department of Study

Electrical & Computer Eng

Discipline

Electrical & Computer Engr

Degree Granting Institution

University of Illinois Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Machine learning
• Mechanistic modeling
• Reinforcement learning
• Disease progression modeling
• Systems biology
• Primary sclerosing cholangitis
• Cholestatic liver disease
• Gut–liver axis
• Bile acid metabolism
• Microbiome

Abstract

Primary sclerosing cholangitis (PSC) is a rare, progressive cholestatic liver disease characterized by bile duct inflammation, hepatic fibrosis, and a markedly increased risk of malignancy. Despite its severity, the underlying pathophysiology of PSC remains incompletely understood, and effective medical therapies remain lacking. This dissertation addresses challenges in modeling PSC and related liver diseases through a spectrum of computational methodologies, including data-driven machine learning models, novel hybrid frameworks that integrate mechanistic and learning-based components, and multi-omics data integration. We begin by assessing the limitations of conventional machine learning approaches in predicting complex clinical outcomes such as hospital readmission in cirrhosis. Using a large, multicenter dataset, we show that even advanced models exhibit only modest gains over traditional risk scores. Conversely, when modeling more biologically grounded outcomes, such as the development of cholangiocarcinoma in PSC, predictive performance improves with the integration of multimodal data, particularly targeted bile acid metabolomics. Nonetheless, these predictive models remain limited in interpretability and are not well suited for exploring therapeutic hypotheses. To overcome these challenges, we introduce REinforcement learning-driven adaptive MEtabolism modeling (REMEDI), a hybrid framework that couples mechanistic ordinary differential equation models of bile acid metabolism with reinforcement learning agents that emulate adaptive physiological responses to cholestatic injury. This hybrid model captures both disease progression and physiological compensation, enabling in silico testing of therapeutic hypotheses in an interpretable manner. We further enhance the mechanistic validity of REMEDI through integration of high-resolution gut microbiome and mycobiome data. We performed one of the most comprehensive shotgun metagenomic and mycobiomic analyses in PSC to date and identified profound dysbiosis and a functional deficiency in microbial bile acid deconjugation. By incorporating these microbiome-derived functional parameters into the REMEDI framework, we created an entero-augmented version that captures host–microbiome interactions and demonstrates how dysbiosis may exacerbate bile acid toxicity and liver injury. In sum, this dissertation presents computational models that bridge data-driven machine learning, mechanistic modeling, and multi-omics insights to improve our understanding of PSC and related liver diseases. By moving beyond static prediction toward dynamic, systems-level modeling, this work lays the foundation for interpretable computational frameworks that can guide precision therapeutics in complex liver diseases.

Graduation Semester

2025-12

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/132636

Copyright and License Information

Copyright 2025 Chang Hu

Owning Collections