When Data Meets Dynamics: AI for Systems Medicine

In complex biological systems, the most important dynamics are often the ones we cannot directly measure. AI–Aristotle is a principled framework for discovering missing terms or unknown interactions in mechanistic models by leveraging indirect observations. Rather than relying purely on data-driven prediction, the method infers the mathematical structure of unobserved dynamics from the data that are available, embedding physical and biological constraints into the learning process. Building on this idea, we introduce compartment-model-informed neural networks (cMINNs), which integrate classical compartmental ODE models with physics-informed learning to uncover hidden mechanisms in systems medicine. As an application, we study tumor response under multiple dosing regimens and use cMINNs to infer latent resistance dynamics and time-varying cell death rates that are not directly observable. This approach enables interpretable discovery of biologically meaningful interactions, providing mechanistic insight into treatment resistance while preserving the predictive power of modern AI.