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

Area of Science:

• Pharmacology
• Toxicology
• Computational Chemistry

Background:

• Drug-induced osteotoxicity presents significant risks in clinical practice and drug development.
• Existing toxicity assessment models lack specialized datasets and algorithms for predicting bone toxicity.
• There is a critical need for predictive tools to evaluate skeletal safety during pharmaceutical research.

Purpose of the Study:

• To develop a predictive model and an online platform for drug-induced osteotoxicity.
• To identify key molecular targets and pathways involved in osteotoxicity.
• To create a robust tool for assessing bone health risks associated with chemical compounds.

Main Methods:

• Compiled a dataset of osteotoxic molecules and performed clustering analysis.
• Utilized target prediction, Gene Ontology (GO), and KEGG pathway analyses.
• Developed and evaluated various machine learning models including Transformer, SVM, XGBoost, and graph neural networks (GNNs) with molecular graphs, ViT, and KPGT.

Main Results:

• Identified key genes (IL6, TNF, ESR1, MAPK3) and elucidated underlying mechanisms.
• Descriptor-based models achieved 0.82 accuracy and 0.89 AUC.
• Molecular graph and KPGT models demonstrated strong predictive performance with AUCs of 0.86.
• Launched Bonetox, the first online platform for osteotoxicity prediction.

Conclusions:

• Bonetox provides a valuable tool for assessing osteotoxicity during drug development.
• The platform aids in improving safety protocols and mitigating skeletal side effects.
• Enhanced prediction of bone toxicity contributes to safer therapeutics and public health.