MCL-DTI: Using Drug Multimodal Informationand Bi-directional Cross-Attention Learningmethod for Predicting Drug-Target Interaction

Abstract Background: Prediction of drug-target interaction (DTI) is an essential step fordrug discovery and drug reposition. Traditional methods are mostlytime-consuming and labor-intensive, and deep learning-based methods addressthese limitations and are applied to engineering. Most of the current deeplearning methods employ representation learning of unimodal information such as SMILES sequences, molecular graphs, or molecular images of drugs. In addition,most methods focus on feature extraction from drug and target alone withoutfusion learning from drug-target interacting parties, which may lead toinsufficient feature representation. Results: To enhance feature learning between drugs and targets, we propose anovel model based on deep learning for DTI task called MCL-DTI which usesmultimodal information of drug and learn the representation of drug-targetinteraction for drug-target prediction. In order to further explore a morecomprehensive representation of drug features, this paper first exploits twomultimodal information of drugs, molecular image and chemical text, to representthe drug. We also introduce to use bi-rectional multi-head corss attention (MCA)method to learn the interrelationships between drugs and targets. Thus, we buildtwo decoders, which include an multi-head self attention (MSA) block and anMCA block, for cross-information learning. We use a decoder for the drug andtarget separately to obtain the interaction feature maps. Finally, we feed thesefeature maps generated by decoders into a fusion block for feature extraction andoutput the prediction results. Conclusions: MCL-DTI achieves the best results in all the three datasets:Human, C.elegans and Davis, including the balanced datasets and an unbalanceddataset. The results on the drug-drug interaction (DDI) task show that MCL-DTIhas a strong generalization capability and can be easily applied to other tasks.