Introduction
Drug-drug interactions (DDIs) account for over 30% of all adverse drug reactions (ADRs) cases and often occur when a patient takes two or more drugs. Many DDIs have been reported, but the underlying mechanisms are poorly understood. Predicting potential DDIs is beneficial to reduce side effects and improve drug safety. Here we present our server, DDI-GCN, to predict DDIs based on inputs of molecular structures.
DDI-GCN utilizes Graph Convolutional Networks (GCN) and co-attention neural networks to identify DDIs and visualize substructures that may be associated with DDIs. Besides, DDI-GCN predicts DDI types that are related to DDI pharmacology (see workflow). A user can submit the structures of a pair of molecules for DDI prediction. We accept molecule input formats of SMILES and drug names.
Given a molecular pair, DDI-GCN can predict:
- DDI probability
- DDI pharmacological type
- Substructures associated with DDIs
- Literature evidences (if available)
- Similarity metrics of targets, enzymes, transporters and side effects between the two input molecules
The prediction model achieved the state-of-the-art performance on independent test. We believe the server can not only help to predict DDI, but also highlight potentially associated substructures which may assist understanding DDI mechanisms and enhancing drug safety.
