Overview

Modern imaging now allows us to observe living systems across molecular, cellular, and tissue scales with unprecedented precision, yet our ability to extract mechanistic understanding from these data remains limited. Unlike molecular omics data, imaging captures continuous high-dimensional spatiotemporal information (e.g., shape, motion) that is inherently complex to represent, analyze, and interpret. There is a deluge of such data, but it remains a challenge to transform such data into interpretable quantitative models and to link them to multimodal data (proteomics, genomics, metabolomics) to reveal the governing principles of biological organization across scales. This workshop will address that challenge by bringing together mathematicians, computer scientists, physicists, and biologists to accelerate the development of image-based scientific machine learning (ML) for biological dynamics — methods capable of learning “laws” of biology directly from experimental images and movies and yielding new insight into how complex forms and behaviors emerge from subcellular structures to whole organs.

​

Several interconnected goals will be pursued: developing physics-informed ML frameworks that embed physical priors such as conservation laws, symmetries, and force balance directly into learning from imaging data; inferring dynamical rules and governing equations from spatiotemporal observations; building principled approaches to bridge scales from subcellular mechanics to tissue and organ morphogenesis; defining cell states and understanding their transitions — how cells move through state space, commit to new identities, and undergo rare but important events such as differentiation decisions and symmetry-breaking; and integrating imaging data with single-cell genomics and spatial transcriptomics to link geometry, mechanics, and motion with molecular state. Practically, the workshop will work toward shared benchmarks, open tools, and community datasets that allow rigorous comparison of new methods and lower the barrier for experimental biologists to use them. Underlying all of this is a meta-goal: cultivating a shared cross-disciplinary language and problem formulation that allows researchers from diverse communities to collaborate productively, laying the foundation for a new era of ML-enabled scientific discovery and theory-building in biology.