A Deep Learning Approach Based on Feature Reconstruction and Multi-dimensional Attention Mechanism for Drug-Drug Interaction Prediction

Drug-drug interactions occur when two or more drugs are taken simultaneously or successively. Early discovery of drug-drug interactions can effectively prevent medical accidents and reduce medical costs. There are already many methods to discover drug-drug interactions. However, the current methods still have much space for performance improvement. We propose a new deep learning approach named FM-DDI based on feature reconstruction and multi-dimensional attention mechanism for drug-drug interactions prediction. The feature reconstruction extracts low-dimensional but informative vector representations of features for the drug from heterogeneous data sources, which can prevent information loss. The deep neural network model based on multi-dimensional attention mechanism gives high weight to critical feature dimensions, which can effectively learn critical information. FM-DDI achieves substantial performance improvement over several state-of-the-art methods for drug-drug interaction prediction. The results indicate that FM-DDI can provide a valuable tool for extracting and learning drug features to predict new drug-drug interactions.