基于图自编码器和协同训练预测miRNA 与疾病的关联.

In rcccnL years, increasing biological cxpcrimcnLs have shown LhaL microRMA (miRNA) plays an imporLanL role in Lhc dcvclopmcnL of human complex diseases・ Thcfclbrc, prcdicLing miRNA・disca8C associaLions can conLribnLc Lo accLiraLc diagnosis and cfTccLivc LrcaLmcnL of diseases・ Since LradiLional biological cxpcrimcnLs arc expensive and Lime-cons Liming, plcnLy of compnLaLional models based on biological daLa have been proposed Lo predieL MiRM A-disease assoc iaLions. In Lhis sLudy, we propose an end-Lo-end deep learning; model Lo predieL miRNA-disease associaLions (MDAGAC)・ Specifically, we fit'sLly consLrucL Lhc similariLv neLwork of miRNA and disease by inLegraLing disease scmanLic similatiLv, miRNA fLincLional similariLv and (yaussian inLcracLion profile kernel similariLv・ Then, Lhc cfTccL of label propagaLion is improved Lhrough Gmph AnLocncodci's and CollaboraLivc Lraining・ This model implcmcnLs Lwo graph anLocncodors on miRNA graph and disease graph rcspccLively, and Lrains Lhese Lwo graph auLocncodcrs collaboraLivcly. Graph auLocncodcrs on miRNA graph and disease graph arc able Lo reconsLrucL score maLrix Lhrough iniLial associaLion maLrix, which is cqnivalcnL Lo pmpagaLc labels on graphs・ The prcdicLion pmbabiliLy of MiRMA-disease associaLion can be obLained from Lhc score maLrix. The rcsulLs of Lhc experimenL based on 5-fold cross validaLion show LhaL MDAGAC is reliable and cffccLivc and oiiLpcrforms currcnL MiRM A-disease associaLions prcdicLion meLhods. [ABSTRACT FROM AUTHOR]