Single-cell morphological profiling reveals insights into cell death

Analysis of single-cell data has emerged as a powerful tool for studying biological processes and response to perturbations. However, its application in morphological profiling is less explored. In this study we profile six cell death subtypes induced by 50 small molecule drugs across six concentrations using the Cell Painting assay. We evaluate the performance of three feature extraction methods at single-cell and aggregated level and apply supervised and unsupervised analyses to uncover factors contributing to cell death mechanisms. Our results show that a bagged LightGBMXT model, trained on single-cell DeepProfiler achieved classification accuracy of 77.23%, with a top overall performance of 89.97% for corresponding aggregated profiles. Furthermore, self-supervised learning using the transformer-based DINO network revealed highly resolved and biologically meaningful subpopulations, shedding light on perturbation- and concentration-specific molecular targets and dose-dependant morphological changes. Our findings demonstrate the potential of studying phenomic data on single-cell level to enhance the characterization of cell death pathways, advancing our understanding of perturbation effects at a granular level.