Unveiling drug-induced osteotoxicity: A machine learning approach and webserver.

Drug-induced osteotoxicity refers to the harmful effects certain pharmaceuticals have on the skeletal system, posing significant safety risks. These toxic effects are critical concerns in clinical practice, drug development, and environmental management. However, current toxicity assessment models lack specialized datasets and algorithms specifically designed to predict osteotoxicity In this study, we compiled a dataset of osteotoxic molecules and used clustering analysis to classify them into four distinct groups Furthermore, target prediction identified key genes (IL6, TNF, ESR1, and MAPK3), while GO and KEGG analyses were employed to explore the complex underlying mechanisms Additionally, we developed prediction models based on molecular fingerprints and descriptors. We further advanced our approach by incorporating models such as Transformer, SVM, XGBoost, and molecular graphs integrated with Weave GNN, ViT, and a pre-trained KPGT model. Specifically, the descriptor-based model achieved an accuracy of 0.82 and an AUC of 0.89; the molecular graph model reached an accuracy of 0.84 and an AUC of 0.86; and the KPGT model attained both an accuracy and an AUC of 0.86. These findings led to the creation of Bonetox, the first online platform specifically designed for predicting osteotoxicity. This tool aids in assessing the impact of hazardous substances on bone health during drug development, thereby improving safety protocols, mitigating skeletal side effects, and ultimately enhancing therapeutic outcomes and public safety.