Abstract:

Cancer exhibits substantial heterogeneity, manifesting as morphological and molecular variation both across and within tumors. This heterogeneity frequently undermines the efficacy of conventional oncological treatments, motivating the development of precision oncology approaches. It arises from both cell-intrinsic drivers and extrinsic factors in the tumor microenvironment (TME), and disease progression is shaped by the dynamic interplay between them, leading patients to follow heterogeneous disease trajectories.

This thesis develops multiscale mechanistic models to delineate the role of heterogeneity in cancer progression and treatment response. Its central approach is a tissue-scale modeling framework used to explore how the TME actively modifies the outcomes of cell-intrinsic disease drivers in castration-resistant prostate cancer.

A secondary aim is to investigate how data-driven approaches can address the costs of mechanistic modeling, through two use cases. Where the mechanism is known and the obstacle is computational, machine learning surrogates trained on simulation output make global sensitivity analysis and feature ranking tractable for complex models. Where the mechanism is only partly known, the same integration of data-driven and mechanistic modeling can instead be turned toward discovering the missing biology: we use scientific machine learning to design a framework for mechanism discovery in a T-cell engager quantitative systems pharmacology (QSP) model.

Zoom Information:

Meeting ID: 236 491 0673
Passcode: 8745