Research On Prediction Of Disease And Drug Resistance Related MicroRNA Based On Neural Network

MicroRNA(mircoRNA)is a conserved non-codingRNA with a length of about22 nucleotides.Although it cannot participate in protein coding,it plays an important role in many human life activities.Studies have shown that the abnormal expression of miRNA is closely related to complex human diseases,and miRNA can also affect drug resistance.Revealing the interaction of miRNA with disease and miRNA with drug resistance is helpful to understand the pathogenesis of disease and the process of drug resistance,which is of great significance for early diagnosis of disease and control of drug efficacy.However,it takes a lot of resources to identify miRNAs related to diseases and drug resistance through biological experiments,and it is blind to a certain extent.Therefore,the development of reliable computational models to identify miRNAs related to diseases and drug resistance can provide guidance for biological experiments.This paper is based on the neural network framework model to identify miRNAs related to potential diseases and drug resistance.Aiming at the problem of predicting disease-related miRNAs,we construct a residual recurrent convolutional neural network model(RCNNMDA)based on the Siamese network.First,we obatain miRNA similarity data and disease similarity data as the characteristic data of miRNA and disease respectively.Secondly,in order to reduce the impact of data noise,the residual recurrent convolutional neural network is used to capture the low-dimensional potential feature representations of miRNAdisease pairs impling efficient bioinformatics feature data,which improves the prediction accuracy.In the model performance evaluation,RCNNMDA achieved0.9292,0.8602,and 0.9097 +/-0.0033 in global leave-one-out cross validation(LOOCV),local LOOCV,and 5-fold cross validation,respectively.In addition,in the case studies of three different diseases,49,50,and 48 of the top 50 miRNAs predicted by RCNNMDA have been verified to be related to the diseases studied.We propose the residual recurrent convolutional neural network to predict drug resistance-related miRNAs(RCNNMRA)and drug resistance-related miRNAs under disease conditions(RCNNMRAD).RCNNMRA integrates miRNA similarity data and drug similarity data as the characteristic data of drug resistance-miRNA pairs,and uses residual recurrent convolutional neural networks to predict.The model has a good performance in the 5-fold cross-validation,with an AUC of 0.9459+/-0.0119.Moreover,in the case study of cisplatin,19 of the top 50 miRNAs predicted by RCNNMRA related to drug resistance were verified by experimental literature.In particular,we also explore the drug resistance affected by miRNA in specific situations from the perspective of disease.We introduce disease information,integrate the similarity data to construct the characteristic data of drug resistance-miRNA-disease pairs,and use residual recurrent convolutional neural network to predict.In the five-fold cross-validation,the AUC of this model is 0.9324.