Understanding Evolving Living Tissues Using Geometric Deep Learning

During key biological processes such as embryonic development, organ formation, and tumor invasion, living tissues exhibit complex behaviors arising from the spatiotemporal interactions among individual cells. Identifying the principles that link tissue-level properties to cellular interactions, and accurately predicting the behavior of individual cells within evolving tissues, remain a significant challenge. Here, we leverage geometric deep learning algorithms to infer the collective dynamics of cells from a static snapshot of cell positions. Furthermore, we show predicting cell-cell rearrangements with single-cell resolution during embryogenesis. Our geometric deep learning model offers a powerful tool for understanding the emergent properties of multicellular living systems and the mechanisms that underlie their development and disease progression.