Abstract
In silico toxicity prediction offers the chance of reducing or replacing most animal testing through the integration of large experimental assay datasets with the appropriate computational approaches. The use of Cell Painting to detect various phenotypic changes induced by chemicals is emerging as a powerful technique in toxicity prediction. However, most Cell Painting approaches use cancer cells that are less relevant for many toxicological endpoints, which may limit the usefulness of this data. In this study, a myoblast cell line is used to characterize cellular responses to a panel of 30 known myotoxicants. In place of traditional structural descriptors, here each perturbation is described by a fingerprint of calculated properties, deducted from the intensity, shape, or texture of individual cells. We show that these kinds of descriptors convey information to allow the prediction of the cellular viability and fate of cells in myoblasts and differentiated myotubes of the C2C12 cell line, and the clustering of drugs by their cytotoxicity responses.