KidneyTox_v1.0 enables explainable artificial intelligence prediction of nephrotoxicity in small molecules

Abstract

Abstract Drug-induced nephrotoxicity remains a leading cause of kidney dysfunction, often with severe or even fatal outcomes. Computational approaches, in particular artificial intelligence (AI), offer a promising alternative by providing reliable, cost-effective, and ethically sound tools for assessing drug-induced nephrotoxicity. Thereby, potentially reducing reliance on animal testing. This study was driven by three core objectives: (i) to analyze the chemical space of compounds associated with drug-induced nephrotoxicity, (ii) to construct a robust supervised machine learning (ML) model for classification, followed by a quantitative Read-Across Structure-Activity Relationship (qRASAR) study, and (iii) to develop an open-access, eXplainable AI (XAI) platform named “KidneyTox_v1.0” ( https://kidneytoxv1.streamlit.app/ ) for nephrotoxicity prediction. Beyond providing predictions, “KidneyTox_v1.0” offers interpretability through interactive SHAP-based waterfall plots, enabling both domain experts and non-experts to understand the contribution of molecular descriptors to toxicity outcomes. These modelling analyses will assist chemists in designing less nephrotoxic molecules in the future.