Modeling adverse liver effects of drugs using kNN QSAR method

Adverse drug reactions (ADRs) continue to be a major cause of drug withdrawals both in development and post-marketing. Quantitative Structure Activity Relationship (QSAR) models have been used to predict human ADRs in the heart. While liver ADRs are a major concern for drug safety, there are currently no in silico models for predicting human liver ADRs. The FDA has assembled a database of human liver ADR data on 490 approved drugs. In this study, we construct a QSAR model capable of performing binary classification (active/inactive) for liver ADRs based on chemical structure using the k-nearest neighbor (kNN) method and rigorous external model validation protocols. Models with high sensitivity (>73%) and specificity (>94%) for external test sets were built. Three databases were screened using our models and the predictions were analyzed. We conclude that QSAR modeling of liver ADRs may be useful in screening pre-clinical drug candidates for potential human hepatotoxicity.