Leveraging the Rich Spatiotemporal Features of live cell imaging with Machine Learning and AI

Live cell imaging is a microscopy technique, where scientists can observe dynamics of living cells across time. One such method, known as lattice lightsheet microscopy captures these processes at high spatiotemporal detail. However, current analyses methods do not always capture the complexities of these feature rich datasets. In this talk, I will use an example of programmed cell death, where cells are exposed to different drugs. Using python packages, such as scikit-image and tsfresh, I will demonstrate how we extract morphological and spatiotemporal features of cells. We use these features to train a supervised machine learning (ML) model to predict which drug treatment the cells were exposed to. Furthermore, using explainable AI with Shapley values, we identify key feature combinations that distinguish the cellular response to each drug. This approach enables data-driven hypothesis generation, allowing us to infer underlying phenotypes and correlate them to the biological processes.