CNNEMS: Using Convolutional Neural Networks to Predict Drug-Target Interactions by Combining Protein Evolution and Molecular Structures Information

Emerging evidences shown that drug-target interactions (DTIs) recognition is the basis of drug research and development and plays an important role in the treatment of diseases. However, the recognition of interactions among drugs and targets by traditional biological experiments is usually blind, time-consuming, and has a high false negative rate. Therefore, it is urgent to use computer simulation to predict DTIs to help narrow the scope of biological experiments and improve the accuracy of identification. In this study, we propose a deep learning-based model called CNNEMS for predicting potential interrelationship among target proteins and drug molecules. This method first uses the Convolutional Neural Network (CNN) algorithm to deeply excavate the features contained in the target protein sequence information and the drug molecule fingerprint information, and then the Extreme Learning Machine (ELM) is used to predict the interrelationship among them. In experiments, we use 5-fold cross-validation method to verify the performance of CNNEMS on the benchmark datasets, including enzymes, ion channels, G-protein-coupled receptors (GPCRs) and nuclear receptors. The cross-validation experimental results show that CNNEMS achieved 94.19%, 90.95%, 87.95% and 86.11% prediction accuracy in these four datasets, respectively. These prominent experimental results indicate that CNNEMS as a useful tool can effectively predict potential drug-target interactions and provide promising target protein candidates for drug research.