We will have a meeting on Friday (March 27) at 3pm in SAS 5270 with speaker Ally Introne. Title and abstract are below.
Title: Inferring Intercellular Interaction Rules for HER2+ Breast Cancer Cell Motility Using Equation Learning
Abstract: Metastasis is a major determinant of survival and treatment efficacy in cancer, yet the mechanisms by which the competition and interaction of heterogeneous tumor cell clones lead to metastasis remain poorly understood. Prior experiments comparing fluorescently barcoded models of human HER2+ breast cancer show that wild-type, d16, and p95 isoforms differ in their invasion and metastatic properties. Here, I present a computational pipeline to better understand the motility features underlying these differences, connecting imaging data to interpretable models of cell–cell interactions. Generative AI–based methods are first used to extract single-cell trajectories from mixed-culture in vitro live-cell timelapse microscopy videos, enabling quantification of motility and spatial correlations. Linear mixed-effects models are then fit, revealing that although the p95 isoform was neither the fastest nor most persistent, it exhibited spatially isolating behavior, suggesting a potential selective advantage not driven solely by increased growth or speed. Because individual cell trajectories are not independent, our statistical analysis required aggregating rich single-cell trajectory data into population-level metrics. This limitation motivates the development of an agent-based model (ABM) that incorporates local interactions and implementation of the Weak Sparse Identification of Nonlinear Dynamics (WSINDy) approach to infer the effects of neighboring cells on motility from the single-cell tracking data. We aim to simulate data from an ABM that recapitulates the motility characteristics of mixed-isoform in vitro HER2+ breast cancer cell populations and then apply the WSINDy approach to test whether intercellular interactions governing cell movement can be recovered from simulated data, and the degree to which recovery accuracy depends on ABM initial conditions such as initial cell density and the proportion of cells within each isoform subpopulation (wild-type, d16, p95).