Natacha Cerisier1*, Tahreem Abbasi1, Marlène Wedler2, Shu Liu3 and Olivier Taboureau1
1Université Paris Cité, INSERM U1133, CNRS UMR 8251, Paris, France
2German Centre for the Protection of Laboratory Animals (Bf3R) and Experimental Toxicology, German Federal Institute for Risk Assessment (BfR)
3German Centre for the Protection of Laboratory Animals (Bf3R) and Experimental Toxicology, German Federal Institute for Risk Assessment (BfR), Berlin, Germany.
natacha.cerisier [at] inserm.fr
Abstract
Drug-induced kidney injury is a leading cause of clinical trial attrition, highlighting the urgent need for early, reliable nephrotoxicity prediction tools. Current in vitro and in silico approaches often fell short in capturing the complex cellular responses underlying renal toxicity; a gap that morphological profiling may help bridge.
Here, we leveraged Cell Painting PLUS, a high-content image-based profiling assay, to investigate whether compound-induced morphological perturbations can serve as predictive signatures of nephrotoxic potential. By integrating nephrotoxicity annotations from multiple curated sources with rich Cell Painting PLUS profiles, we explored a novel application of phenotypic profiling beyond its traditional use in mechanism-of-action studies.
Starting from a raw dataset, rigorous feature selection and compound curation pipelines were applied to maximize data quality, leading to 90 annotated compounds. Unsupervised exploration via hierarchically-clustered heatmaps (clusterMap) revealed structured patterns distinguishing nephrotoxic from non-nephrotoxic compounds, and identified morphological features driving this separation.
Building on these insights, we trained and benchmarked several machine learning classifiers. A Random Forest model achieved the best predictive performance, with an AUC of 82%, a sensitivity of 87%, and a specificity of 63%. These results provided a proof of concept that high-content image-based profiles encode biologically meaningful signals relevant to nephrotoxicity — and laid the groundwork for scalable, image-driven toxicity screening pipelines.
Keywords: Nephrotoxicity, Cell-Painting, Machine-learning, high-content-imaging
Acknowledgement: This project receives funding from the European Union’s Horizon 2020 Research and Innovation program under Grant Agreement No. 964537 (RISK-HUNT3R), and it is part of the ASPIS cluster.