Prediction Of LncRNA-Protein Interaction And LncRNA Function Based On Deep Learning

Lnc RNA refers to non-coding RNA with a length of more than 200 bases.They regulate gene expression at the epigenetic,transcriptional,and post transcriptional levels.A large amount of evidence indicates that lncRNA plays a pivotal role in many life activities by interacting with proteins.In addition,current functional prediction of lncRNAs mostly relies on lncRNA-protein interactions.Therefore,identifying the interaction between lncRNA and proteins to further predict the function of lncRNA is crucial for understanding the formation mechanism and potential applications of lncRNA.Traditional wet biological experimental methods are costly,complex,and time-consuming,making it difficult to quickly identify lncRNA protein interactions and predict lncRNA function.In recent years,many computer-based methods have been applied to the study of lncRNA,but there are problems such as single features,insufficient interpretability,or low prediction accuracy.Therefore,this article has conducted in-depth research on lncRNA-protein interaction and prediction of lncRNA function based on in-depth learning methods.The main work is as follows:(1)A new method for predicting lncRNA protein interaction relationship,DFRPI,has been proposed.Firstly,the initial features of lncRNA protein interaction are extracted from the sequence and structure of lncRNA protein.Secondly,use a deep automatic encoder to learn appropriate encoding parameters and construct the best descriptive features based on the initial features.Then,the marginal Fisher analysis algorithm is used to optimize feature encoding parameters and define the optimal classification features for lncRNA protein interactions.Finally,a prediction model based on random forest is constructed by using the optimal classification features to predict the lncRNA protein interaction relationship.The experimental results show that the accuracy,recall,specificity,sensitivity,accuracy,Matthew correlation coefficient,and area under the ROC curve of this method reach 0.920,0.916,0.920,0.916,0.918,0.836,and0.906,respectively,and the overall performance is superior to the existing methods.(2)A new method for predicting lncRNA function has been proposed.First,a random walk algorithm accompanied by a restart is executed on the lncRNA to lncRNA similarity network established from lncRNA co expression data to generate low dimensional topological features of lncRNA.Secondly,the neighbor counting method is used to extract the semantic features of lncRNA according to the lncRNA protein interaction relationship and the Gene Ontology(GO)information of the protein.Then,the semantic and topological features of lncRNA are fully fused using a graph convolutional neural network to obtain the GO functional features of lncRNA.Finally,a typical correlation autoencoder(C2AE)was used as a classifier to achieve functional prediction of lncRNA.This method was compared with existing lncRNA function prediction methods on the independent test set of lncRNA2GO-55.The experimental results showed that the accuracy,recall,and F1 score of this method reached 0.274,0.683,and 0.397,respectively,and correctly labeled the functions of 54 lncRNAs in the dataset lncRNA2GO-55 containing a total of 55 lncRNAs,making it the best performing method among all methods.