Existing approaches to cellular morphological characterization often rely on classification frameworks that require excessive training, are system-specific and challenged by cell-to-cell heterogeneity. Machine learning based methods, while powerful, typically operate as a black-box system, offering limited interpretability and minimal insight into the underlying morphological alterations. We present cell morphology fingerprinting, a versatile framework that robustly captures morphological changes of cellular compartments at the single-cell level. Our approach extracts over 40 key morphological features related to the intensity, geometry, and texture of the nucleus and cytosol; that are both interpretable and biologically meaningful, providing a direct link between perturbations and their cellular consequences. This method is broadly applicable across diverse cellular systems for quantitative characterization of cellular morphology and perturbations, including chemical agents, drugs, genetic modifications, physical stimuli, and stress responses. To demonstrate its utility, we applied morphology fingerprinting to multiple morphologically diverse cell lines treated with 8 different chemical compounds and drugs with distinct mode of actions. Using a simple classifier with feature importance, the model accurately and rapidly identified distinct and compound-specific morphological signatures and predicted the drug treatment based on the morphological feature variation. Importantly the method allowed for direct quantification of the degree of response of each individual cell to each drug. This work establishes cell morphology fingerprinting as a robust and accessible methodology for correlating functional cell phenotypes with morphological features at the cellular level, offering broad applications in drug screening, cellular physiology, and mechanistic studies. The provided user-friendly interface will ease its broad adoption by the community.
Bohr et al. (Sun,) studied this question.