Research On Prediction Of Disease-associated Non-coding RNAs And Microbes Based On Heterogeneous Network

More and more evidence shows that non-codingRNAs and microbes play an important role in the occurrence and development of human diseases.Traditional biological experimental methods to identify non-codingRNAs and microbes related to diseases have disadvantages such as costly and time-consuming.Therefore,it is urgent to effectively predict the potential non-codingRNAs and microbes related to diseases by computational methods,so as to reduce the biological experimental cycle,number of experiments and experimental costs.In recent years,although important results have been achieved in the research on prediction of disease-related non-codingRNAs and microbes,there is still room for further improvement.In this paper,prediction of disease-related non-codingRNAs and microbes were studied in depth.The main research contents of this paper are as follows:(1)In the research of predicting non-codingRNA-disease associations,this paper mainly studies the problem of miRNA-disease and circRNA-disease associations prediction.Based on the existing methods for predicting miRNA-disease associations,this paper proposed a novel method named PMDASNF for predicting miRNA-disease associations.In this method similarity network fusion algorithm to fuse multiple biological information networks to improve the prediction performance of the model.Based on leave-one-out crossvalidation and 5-fold cross-validation,the AUC achieved by PMDASNF were 0.9400 and 0.9144±0.0009,respectively.Based on the existing methods for predicting circRNA-disease associations,this paper proposed a novel method named IGCHCDA for predicting circRNAdisease associations.In this method,linear network fusion to integrate Gaussian interaction profile kernel similarity and cosine similarity to improve the prediction performance of the model.Based on leave-one-out cross-validation and 5-fold cross-validation,the AUC achieved by IGCHCDA were 0.8807 and 0.8631±0.0059,respectively.The experimental results show that PMDASNF and IGCHCDA have better effects in predicting miRNAdisease and circRNA-disease associations,respectively.(2)In the research of predicting microbe-disease associations,a new method called Pred HMDA was proposed to predict microbe-disease associations in this paper.In the Pred HMDA,the Gaussian interaction profile kernel similarity and cosine similarity were fused for the first time,and the network consistency projection was used to complete the prediction of microbe-disease associations.On the benchmark data set,based on leave-oneout cross-validation and 5-fold cross-validation,the AUC reached by Pred HMDA were 0.9589 and 0.9361±0.0037,respectively.To further demonstrate the effectiveness of the Pred HMDA,case studies of asthma,colorectal cancer,and inflammatory bowel disease were conducted,where 10,8 and 10 of the top 10 predicted microbes were confirmed by relevant literatures.The code and data set are available at http://github.com/August Me/Pred HMDA.The three methods of PMDASNF,IGCHCDA and Pred HMDA proposed in this paper have achieved good performances on miRNA-disease,circRNA-disease,and microbedisease associations prediction,respectively,which provide certain help for the exploration of the pathogenesis of diseases and the diagnosis,treatment and prognosis of diseases.