Revealing invisible cell phenotypes with conditional generative modeling
Biological sciences, drug discovery and medicine rely heavily on cell phenotype perturbation and observation. Aside from dramatic events such as cell division or cell death, most cell phenotypic changes that keep cells alive are subtle and thus hidden from us by natural cell variability: two cells in the same condition already look different. While we show that deep learning models can leverage invisible features from microscopy images, to discriminate between close conditions, these features can yet hardly be observed and therefore interpreted. In this work, we show that conditional generative models can be used to transform an image of cells from any one condition to another, thus canceling cell variability. We visually and quantitatively validate that the principle of synthetic cell perturbation works on discernible cases such as high concentration drug treatments, nuclear translocation and golgi apparatus assays. We then illustrate its effectiveness in displaying otherwise invisible cell phenotypes triggered by blood cells under parasite infection, the presence of a disease-causing pathological mutation in differentiated neurons derived from iPSCs or low concentration drug treatments. The proposed approach, easy to use and robust, opens the door to the accessible discovery of biological and disease biomarkers.