High throughput method for accurate prediction of compound-induced liver injury

Abstract

A method and system for predicting liver injury in vivo due to hepatocyte damage by a test compound are provided. The method includes acquiring images of fluorescently stained cells obtained from a cell culture in which the cells have been treated with a dose-range of at least the test compound and its vehicle. The cells may be hepatic cells including primary or immortalized hepatocytes, hepatoma cells or induced pluripotent stem cell-derived hepatocyte-like cells. The acquired images are segmented. The method further includes extracting and analyzing one or more phenotypic features from the segmented images, wherein the one or more phenotypic features are selected from the group of intensity, textural, morphological, or ratiometric features consisting of (a) features of DNA, (b) features of RELA (NF-KB p65), and (c) features of actin filaments at different subcellular regions and d) features of cellular organelles and their substructures in the segmented images. Finally, the method includes normalizing results from the treated samples to vehicle controls and predicting the probability of liver injury by the test compound based on test compound-induced normalized changes of the extracted and selected phenotypic features using machine learning methods.