Sheehan S, et al. - Via a cohort of mice with different genotypes, the researchers investigated histological images of Periodic acid-Schiff–stained renal sections by deep neural networks (DNN)-based transfer learning and demonstrated strong generalization performance of DNN-based machine on multiple histological image processing tasks. Superior performance at segmenting glomeruli from the non-glomerular structure was noticed further, predicting the genotype of the animal based on glomerular quantitative image features. Correlation between DNN-based genotype classifications and mesangial matrix expansion was found. Through analysis of non-glomeruli images, the neural network could recognize novel histological features that differed by genotype, including the vacuoles presence, nuclear count, and integrity of proximal tubule brush border, which was certified via immunohistological staining. Hence, the power of DNNs in the extraction of biologically relevant phenotypes and in serving as a platform for identification of novel phenotypes was exhibited highlighting the synergistic potentialities for pathologists and DNNs for essentially scaling up their ability to generate novel mechanistic hypotheses in disease.