LncRNA And Disease Association Prediction Method Research

The function of lncRNAs almost involve the whole biological process of biological physiology and pathology.The researches on the role of the lncRNAs in the pathological process help reveal their mechanism of action and provide new ideas or solutions for the prevention,diagnosis and treatment of the diseases.Identifying the associations between lncRNAs and diseases is an important basis for determining the function of lncRNAs in the process of diseases.The traditional methods for predicting lncRNA-disease associations are based on shallow learning methods which cannot learn the deep and complex representations of lncRNA-disease associations,and seldom consider the different importance of various information.In this paper,two prediction methods of lncRNA-disease associations are proposed.The first association prediction method is based on convolutional neural networks and long short-term memory neural networks.The second one is based on a dual convolutional neural networks with attention mechanisms.The first method is a prediction model of multi-model fusion,which is based on the convolution neural networks prediction model and the long short-term memory neural networks prediction model.The prediction model based on convolutional neural networks firstly constructs the feature matrix from diverse biological premises about lncRNAs,miRNAs,and diseases,and then learns the original and global information of lncRNA and disease associations.The neural networks prediction model based on long short-term memory first explores the topology of the lncRNA-miRNA-disease networks through the random walk,and then constructs the feature vector by combining the original information and the timing information of the random walk,and finally considers the inconsistency in the timing and the different importance of features.The results of evaluations and case studies show that this method is superior to the traditional method.The second method is based on a dual convolutional neural networks with attention mechanisms which has three modules.The convolution module learns the original and global information representations of lncRNA-disease associations.The attention module considers the different effects of the feature level and the relationship level on the association predictions and learns the attention representations through the two-layer attention mechanism and the convolutional neural networks.The convolution and full connection module considers the combination of dual convolutional neural network information and the deeper representations,and gives predicted resuls of associations.Different experimental results and case studies show that this method has the best performance.