Adverse Event Classification from Co-prescribed Drugs by Integrating Chemical, Phenotypic and Graph Embedding Features

Adverse drug reaction prediction is important before releasing the drug into markets. It is one of the significant causes of failure in drug progression in the pharmaceutical industry. A post-marketing undetected adverse event can lead to severe health conditions or morbidity. This paper proposes a technique for prediction of drug-drug-adverse event. The prediction model learns chemical, phenotypic, and graph embedding features. We explored both machine learning and deep learning approaches. The model predicts the presence of adverse event for the co-prescribed drug with an Area Under Curve (AUC) score of 0.90 and the accuracy with the 0.83.