Regensburg 2025 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

BP: Fachverband Biologische Physik

BP 32: Computational Biophysics II

BP 32.13: Talk

Friday, March 21, 2025, 12:45–13:00, H46

Ab-initio optimization and AI-powered inference for parametrizing complex biological models under low data availability — •Thomas R. Sokolowski — Frankfurt Institute for Advanced Studies (FIAS), Ruth-Moufang-Str. 1, 60438 Frankfurt am Main, Germany

Early development unfolds under diverse circumstances and time scales, but always facing the impacts of inevitable biological noise. To cope with this, various developmental mechanisms evolved, with their differences shaped by physical and environmental constraints. In spite of decades of research, we still lack theories that explain these processes truly mechanistically. Increasing computational power allows for constructing developmental models with increasing complexity, but since corresponding experimental data is scarce, the parametrization of such models becomes a key problem itself. I will contrast two strategies for parametrizing biophysical models in development and beyond: optimization of normative theories, and Bayesian inference. I will present a framework that unifies both strategies in a mathematically rigorous fashion and enables quantitative transition between them. I will then present our results combining both strategies for understanding embryogenesis in two organisms: (1.) optimization of a spatial-stochastic model of the gap gene system in Drosophila, and (2.) elucidation of robust cell-fate assignment in early mouse embryogenesis via AI-powered simulation-based inference (SBI). Our results highlight distinct developmental strategies that emerged under the different circumstances faced by the two organisms.

Keywords: Embryo development; stochastic simulation; simulation-based inference; optimization; information theory