Machine learning predicts adverse drug reactions based on drugs’ in vitro pharmacology

In an effort to minimize adverse drug reactions, researchers at Harvard Medical School, the Jackson Laboratory, the Massachusetts Institute of Technology, the University of Massachusetts, Northeastern University, Oracle Health Sciences, and the Novartis Institutes for Biomedical Research studied adverse drug reaction events from post-marketing identification surveys and target-based in vitro pharmacology of more than 2,000 marketed drugs.

Through machine learning, they were able to systematically predict the drug effects on human patient populations from their target-based preclinical profiles. The work was published online in EBioMedicine on June 17.

The team validated the machine learning predictions based on chronological event reporting, comparison with drug labels, and systematic text mining of scientific literature. Using a target-centric approach, they identified 221 statistical associations between protein targets and adverse reactions, providing novel insight into the molecular components underlying physiological adverse reactions.

The machine learning algorithms, which can be downloaded online, are scalable and adaptable to similar datasets.