From trajectories to models: data-driven approaches to decipher the dynamics of living systems

Resumo: Stochastic differential equations are often used to model the dynamics of living systems, from Brownian motion at the molecular scale to the dynamics of cells and animals. How does one learn such models from experimental data? This task faces multiple challenges, from information-theoretical limitations to practical considerations. I will present a recent and ongoing effort to develop new methods to reconstruct such stochastic dynamical models from experimental data, with a focus on robustness and data efficiency. These methods provide a generic means to quantify complex behavior and unfold the underlying mechanisms of an apparently erratic trajectory.