Drug toxicity prediction based on genotype-phenotype differences between preclinical models and humans.

Area of Science:

• Pharmacology and Toxicology
• Computational Biology
• Drug Development

Background:

• Drug development faces challenges due to poor translatability of preclinical toxicity findings to human outcomes.
• Biological differences between humans and model organisms cause high clinical trial attrition and drug withdrawals.
• Current toxicity prediction methods often overlook inter-species differences, focusing mainly on chemical properties.

Purpose of the Study:

• To develop a machine learning framework for predicting human drug toxicity by incorporating genotype-phenotype differences (GPD).
• To improve the accuracy of toxicity predictions by accounting for biological variations between preclinical models and humans.
• To reduce drug development costs and enhance patient safety through early identification of high-risk drugs.

Main Methods:

• Developed a machine learning framework integrating GPD between preclinical models (cell lines, mice) and humans.
• Assessed GPD of drug targets across gene essentiality, tissue expression, and network connectivity.
• Benchmarked the GPD-based model against state-of-the-art predictors using independent datasets and chronological validation.

Main Results:

• GPD features were significantly associated with drug failures due to severe adverse events in a dataset of 434 risky and 790 approved drugs.
• The Random Forest model integrating GPD and chemical features showed enhanced predictive accuracy (AUPRC = 0.63, AUROC = 0.75).
• The model demonstrated superior performance for neurotoxicity and cardiovascular toxicity, outperforming chemical structure-based models and anticipating real-world drug withdrawals.

Conclusions:

• Incorporating genotype-phenotype relationships provides a biologically grounded strategy for drug toxicity prediction.
• The developed framework enables early identification of high-risk drugs, promising reduced development costs and improved patient safety.
• This approach can increase the success rate of therapeutic approvals by enhancing prediction of clinical outcomes.