Geometric deep learning reveals the spatiotemporal features of microscopic motion

Abstract The characterization of dynamical processes in living systems provides important clues for their mechanistic interpretation and link to biological functions. Owing recent advances microscopy techniques, it is now possible routinely record the motion cells, organelles individual molecules at multiple spatiotemporal scales physiological conditions. However, automated analysis dynamics occurring crowded complex environments still lags behind acquisition microscopic image sequences. Here we present a framework based on geometric deep learning that achieves accurate estimation properties various biologically relevant scenarios. This deep-learning approach relies graph neural network enhanced by attention-based components. By processing object features with priors, capable performing tasks, from linking coordinates into trajectories inferring local global dynamic properties. We demonstrate flexibility reliability this applying real simulated data corresponding broad range experiments.