Location
MC 6460
Candidate
Saranya Varakunan | Applied Mathematics, University of Waterloo
Title
Mechanistic Modeling and Scientific Machine Learning for Inference and Model Discovery in Cancer Immunology
Abstract
Cancer arises from complex interactions among biological processes that evolve over time and vary substantially between patients. Mechanistic models provide a framework for studying these dynamics, but immunological mechanisms are often only partially known, and uncertainty in model parameters and structure can restrict the questions that these models can reliably address. This research explores mathematical and machine learning approaches for inverse problems and model discovery under these limitations. Initial work developed a mechanistic model of CD4+ and CD8+ CAR-T cell dynamics to investigate how CAR-T product composition and patient heterogeneity influence treatment response. This framework was complemented by data-driven methods for predicting treatment outcomes when patient-specific quantities are uncertain. A second direction uses Generative Flow Networks (GFlowNets) to learn distributions over plausible governing equations from sparse data generated by a gene regulatory network model; upcoming work will extend this probabilistic perspective to learning distributions over hematopoietic differentiation graphs from single-cell data. Finally, the proposed research will investigate theoretical questions concerning learning and identifiability in inverse problems, drawing on tools from statistical physics and high-dimensional learning.