Learning Pairwise Potentials via Differentiable Recurrent Dynamics

This paper describes a novel approach to optimizing parameters defining the behavior of pairwise potential models, which can simulate cell migration.Determining appropriate parameters remains a major challenge due to measurement difficulties and high computational cost.To solve this problem, we propose a fully differentiable framework that integrates metrics for evaluating cell positional configurations with recurrent intercellular interaction dynamics.The experimental results showed that our method successfully estimated parameters that reproduce cell positional configurations in both synthetic and biological datasets.