Location
MC 6460
Candidate
Morghan Van Walsum | Applied Mathematics, University of Waterloo
Title
Applications of integrated mechanistic modelling and scientific machine learning in mathematical oncology
Abstract
Cancer treatments are evolving from traditional treatments centred around general cytotoxicity to more specialized treatments that target features specific to the hallmark properties of cancer. However, the precise biological processes and molecular dynamics underlying these properties are complex and remain poorly understood. This has limited the efficacy of these novel therapies, and in some cases resulted in unanticipated side effects. Mechanistic modelling, primarily implemented through differential equations, allows for the exploration of these dynamics and their consequences through a theoretical lens, enabling scenario exploration and pattern discovery that is not feasible in a clinical setting. Scientific Machine learning (ML) approaches can be integrated with mechanistic modelling to define and explore poorly understood system dynamics in a data-driven way. In this presentation, I will outline an ODE-ML framework for mechanistic inference of population dynamics during tumour-vasculature recruitment using Universal Physics Informed Neural Networks (UPINNs). I will then discuss relevant extensions of this work, including application to PDE models and modification of the standard UPINN pipeline to account for the possibility of unknown physics existing in the system. Finally, I will present potential applications of mechanistic modelling integrated with ML for several open questions in mathematical oncology, including molecular dynamics of the tumour-immune response and cancer-specific trends in cell metabolism.