PROJECT SUMMARY/ABSTRACT Live cell imaging has revolutionized the study of dynamic cellular processes, enabling researchers to gain valuable functional insights that would be impossible to obtain from static images. The recent advances in microscopy technology have made it possible to acquire an unprecedented amount of live cell image data at high spatiotemporal resolutions. However, this abundance of data has brought new challenges, particularly concerning phenotypic and causal heterogeneity. This heterogeneity can complicate the analysis and interpretation of the data, making it challenging to identify critical mechanistic details and understand how cellular processes function in different conditions. To address this issue, we have been developing machine learning platforms that can deconvolve the heterogeneity of live cell images at subcellular and cellular levels. However, there are still many challenges in analyzing live cell heterogeneity in greater detail. First, conventional feature selection/learning that aids in classifying known phenotypes could eliminate biologically meaningful heterogeneity. Second, current cell biology research focuses primarily on mechanisms that produce similar average effects across various populations, which could suffer from causal heterogeneity. Even with the significant progress made in single-cell biology, the discovery of causal mechanisms at the single-cell resolution is still not fully achieved. Finally, the substantial heterogeneity in cell motility and morphodynamics poses a challenge to obtaining integrated and systematic understandings of these processes, despite their clear significance in fields such as tissue regeneration and cancer metastasis. To overcome these challenges, we propose to advance current machine learning methods to i) identify features that preserve subtype heterogeneity while being discriminative between known phenotypes, ii) acquire causal datasets of live cell images by tracking cells before and after optogenetic treatment, and iii) deconvolve the causal heterogeneity of single live cells by integrating time-series modeling and deep learning. We will apply these methods to map out single-cell mechanisms governing cellular motility and morphodynamics of various cell types, which will be valuable resources for the cell biology community. Then, we will characterize how mechanical and metabolic perturbations can shape the heterogeneous landscapes of cell motility and morphodynamics, providing new insights into the underlying mechanisms integrating these processes.