Bayesian model selection for partially observed second order Langevin dynamics

To describe and elucidate the mechanisms of cell motility, numerous SDE-based Langevin models have been proposed. However, the data-driven comparison between these models has not been actively studied. In this talk, I will present a Bayesian approach to estimate and compare tentative first- and second-order Langevin models. By using the likelihood approximation technique introduced in Ref. [arxiv:2411.08692], we develop a framework that ranks them based on the calculated marginal likelihood. We test and benchmark the approach using synthetic data and subsequently apply it to time-series data of Dicty cells to find the best model for their ameboid motility.