The Study Of Disease And Small Molecule Drug-related MicroRNA Prediction Based On Matrix Decomposition And Heterogeneous Graph Inference

Micro RNA(mi RNA)is a class of short non-coding RNA molecules(about 22 nucleotides in length)encoded by endogenous genes.Studies have found that mi RNA is involved in a series of important processes in life,including cell proliferation,development,differentiation,apoptosis,metabolism,aging,signal transduction and virus transfection.Besides,various complex disease-related mi RNAs have been found,which are helpful for further exploring the molecular mechanism of disease pathogenesis.Furthermore,since mi RNAs are emerging as new biomarkers for disease,the identification of potential disease-related mi RNAs is also helpful for the detection and diagnosis of disease.Meanwhile,studies have also confirmed that small molecule drugs can regulate the expression and function of mi RNA and are thus potential regulators of mi RNA.Hence,mi RNA is gradually becoming a novel high-value target of drug therapy.Identifying potential small molecule drug-related mi RNAs not only helps to understand the relationship between them,but also significantly accelerates the development of mi RNA-targeted drugs.Here,two prediction models were proposed for the above two research problems: one model of MDHGI for disease-related mi RNA prediction based on matrix decomposition and heterogeneous graph inference;another model of SLHGISMMA for small molecule drug-related mi RNA prediction based on matrix decomposition and heterogeneous graph inference.The matrix decomposition algorithm can remove the noise of the original association matrix to a certain extent,while the heterogeneous graph inference can effectively utilize the implicit topological information in the heterogeneous graph(obtained by fusing new association information and similarity information)to make prediction.Therefore,the effective combination of matrix decomposition and heterogeneous graph inference makes MDHGI and SLHGISMMA have good prediction performance.In performance evaluation,MDHGI obtained AUCs of 0.8945,0.8240 and 0.8794+/-0.0021 in global leave-one-out cross validation(LOOCV),local LOOCV and 5-fold cross validation,respectively.In addition,in case studies of esophageal cancer,lymphoma,lung cancer and breast cancer,49,49,50 and 50 of the top 50 disease-related mi RNAs predicted by MDHGI were confirmed by experimental literatures,respectively.As for SLHGISMMA,based on dataset 1(dataset 2),it obtained AUCs of 0.9273(0.7774),0.7703(0.6556),0.9365(0.7973)and 0.9241+/-0.0052(0.7724+/-0.0032)in global LOOCV,small molecule drug-fixed local LOOCV,mi RNA-fixed local LOOCV and 5-fold cross validation,respectively.In addition,in case studies of 5-Fluorouracil,17?-Estradiol,and 5-Aza-2'-deoxycytidine,39,30 and 26 of the top 50 small molecule drug-related mi RNAs predicted by SLHGISMMA were confirmed by database or experimental literature,respectively.Overall,the results in the performance evaluation confirm the reliability and stability of MDHGI and SLHGISMMA.