MultiCell: geometric learning in multicellular development.

During developmental processes such as embryogenesis, how a group of cells self-organizes into specific structures is a central question in biology. However, it remains a major challenge to understand and predict the behavior of every cell within the living tissue over time during such intricate processes. Here we present MultiCell, a geometric deep learning method that can accurately capture the highly convoluted interactions among cells. We demonstrate that multicellular data can be represented with both granular and foam-like physical pictures through a unified graph data structure, considering both cellular interactions and cell junction networks. Using this method, we achieve interpretable four-dimensional morphological sequence alignment and predict single-cell behaviors before they occur at single-cell resolution during Drosophila embryogenesis. Furthermore, using neural activation map and model ablation studies, we demonstrate that cell geometry and cell junction networks are essential features for predicting cell behaviors during morphogenesis. This method sets the stage for data-driven quantitative studies of dynamic multicellular developmental processes at single-cell precision, offering a proof-of-concept pathway toward a unified morphodynamic atlas.