Improved prediction of drug-drug interactions using ensemble deep neural networks

Nowadays, combining multiple drugs is the optimal therapy to decelerate the pathologic process, which contains various underlying adverse effects due to drug-drug interactions (DDIs). Artificial intelligence (AI) has the potential to evaluate the interaction, pharmacodynamics, and possible side effects between drugs. Over the past years, many AI-based DDI prediction techniques, including both machine learning and deep learning, that harness available big data have been presented. Although significant progressives have been gained through previous methods, improvements are still crucial. In this work, we introduce an ensemble deep neural network that can help to improve the predictive performance of DDIs. As a result, our prediction model could predict 86 types of DDIs on a benchmark dataset with an average accuracy of 93.80%. Our ensemble classifier produces better performance when compared with the existing proposed methods on the same dataset. Our model's high performance places it among the top list of those well-built pharmacovigilance-assisted tools that facilitate the detection of DDIs to support medical decisions and drug development. We released our source codes and models at https://github.com/khanhlee/edn-ddi.