Image:
August 15, 2023
We develop a hybrid approach to automate model construction that extends baseline physics models by features selected from statistical learning and testing procedures. Applied to fibroblast cells undergoing a disorder-to-order transition, our approach reveals four additional features: anisotropic and non-Gaussian velocity fluctuations, novel neighbor ensembles, and weights of interactions. Our model can accurately produce the temporal progression of global orientational order parameter and the variability of velocity, whereas models missing any of the selected features lead to large discrepancy. Our approach has broader applications in understanding active matter systems and represents a paradigm shift in modeling their dynamics.