Abstract of the talk// Empirical force fields enable atomistic simulations of complex biological systems by replacing electronic structure with simplified classical interactions. Because of their empirical nature and often ad hoc parametrization, these models are frequently debated and, often for good reason, viewed with skepticism. In this talk, Vojtech will discuss their fundamental limitations and explore physical anchors that can guide their systematic improvement, in particular the reintroduction of electronic polarization through the electronic continuum correction (ECC) and the use of ab initio data for benchmarking in the most physical sense.

Building on this perspective, he will present a Bayesian framework for learning force-field parameters in a probabilistic manner, allowing uncertainties arising from the approximate representation of electronic structure to be quantified explicitly. Finally, he will show how these ideas can be united to address a challenging problem for classical simulations: modeling calcium binding in troponin C, the molecular trigger of calcium-dependent muscle contraction.