MAPTrans: mutual attention transformer with dynamic meta-path pruning for drug repositioning

Drug repositioning has become a hot topic that could provide an innovative solution in drug discovery by exploring the potential correlation between drugs and diseases. However, existing computational drug repositioning methods fail to effectively integrate heterogeneous data from multiple sources and neglect the multi-level and multi-scale interactions in biological systems. To address the above problems, we propose MAPTrans, which dynamically optimizes the representation of disease and drug with a multi-level meta-path aggregation strategy. In addition, a multi-view importance assessment mechanism is introduced to evaluate and filter the most discriminating views to optimize feature representation. A mutual attention mechanism Transformer architecture with a cross-view interaction that fuses the information of drugs and diseases in a multi-view space is designed. Experimental results of MAPTrans on multiple benchmark datasets show that it significantly outperforms existing baseline models.