Research On Drug Relocation Method Based On Deep Learning

Identifying the novel indications for the approved drugs is one of the most important steps in bioinformatics in recent years.Most traditional drug repositioning methods only consider the node information in the drug-disease association network and use matrix decomposition for association prediction;or considering the connection between nodes,the association prediction is performed by random walk.However,these two shallow methods are difficult to express complex biological information.In this case,how to predict drug-disease association through deep learning networks is a challenging task.In this paper,two drug repositioning models based on deep learning networks are proposed.The first model is CBPred,a two-way deep learning models based on bidirectional long-short-term memory networks and convolutional neural networks.CBPred captures the information of self-attributes of the drug-disease node pairs and the topological information of the drug-disease association network.In addition,to distinguish the importance of different paths,we propose a path-level attention mechanism to assign different attention weights for different paths.Compared with other excellent drug repositioning models,the prediction effect of CBPred is better.The second model is a deep learning-based drug repositioning model based on residual network and convolutional long-short-term memory network,named as Res CLPred.Res CLPred express and learn biological data through as complex and deep network structures as possible.This paper uses Adam to automatically update and optimize the network parameters involved in the model.Experimental results on real data show that the model can more accurately recommend reliable drug-disease candidates,and it is superior to existing shallow prediction models in multiple metrics.