Date of Award
12-2023
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
School of Computing
Committee Chair/Advisor
Nina C. Hubig
Committee Member
William J. Richardson
Committee Member
Brian C. Dean
Committee Member
Bethany J. Wolf
Committee Member
Awe J. Schoepf
Abstract
Biochemical and biomechanical signals drive cardiac remodeling, resulting in altered heart physiology and the precursor for several cardiac diseases, the leading cause of death for most racial groups in the USA. Reversing cardiac remodeling requires medication and device-assisted treatment such as Cardiac Resynchronization Therapy (CRT), but current interventions produce highly variable responses from patient to patient. Mechanistic modeling and Machine learning (ML) approaches have the functionality to aid diagnosis and therapy selection using various input features. Moreover, 'Interpretable' machine learning methods have helped make machine learning models fairer and more suited for clinical application. The overarching objective of this doctoral work is to develop computational models that combine an extensive array of clinically measured biochemical and biomechanical variables to enable more accurate identification of heart failure patients prone to respond positively to therapeutic interventions. In the first aim, we built an ensemble ML classification algorithm using previously acquired data from the SMART-AV CRT clinical trial. Our classification algorithm incorporated 26 patient demographic and medical history variables, 12 biomarker variables, and 18 LV functional variables, yielding correct CRT response prediction in 71% of patients. In the second aim, we employed a machine learning-based method to infer the fibrosis-related gene regulatory network from RNA-seq data from the MAGNet cohort of heart failure patients. This network identified significant interactions between transcription factors and cell synthesis outputs related to cardiac fibrosis - a critical driver of heart failure. Novel filtering methods helped us prioritize the most critical regulatory interactions of mechanistic forward simulations. In the third aim, we developed a logic-based model for the mechanistic network of cardiac fibrosis, integrating the gene regulatory network derived from aim two into a previously constructed cardiac fibrosis signaling network model. This integrated model implemented biochemical and biomechanical reactions as ordinary differential equations based on normalized Hill functions. The model elucidated the semi-quantitative behavior of cardiac fibrosis signaling complexity by capturing multi-pathway crosstalk and feedback loops. Perturbation analysis predicted the most critical nodes in the mechanistic model. Patient-specific simulations helped identify which biochemical species highly correlate with clinical measures of patient cardiac function.
Recommended Citation
Haque, Anamul, "Interpretable Mechanistic and Machine Learning Models for Pre-dicting Cardiac Remodeling from Biochemical and Biomechanical Features" (2023). All Dissertations. 3481.
https://open.clemson.edu/all_dissertations/3481
Author ORCID Identifier
0000-0002-4919-127X
Included in
Bioinformatics Commons, Biomedical Informatics Commons, Cardiovascular Diseases Commons, Systems Biology Commons