A machine learning framework that integrates multi-omics data predicts cancer-related LncRNAs

Abstract Background LncRNAs (Long non-coding RNAs) are a type of RNA molecule with transcript length longer than 200 nucleotides. LncRNA has been novel candidate biomarkers in cancer diagnosis and prognosis. However, it is difficult to discover the true association mechanism between lncRNAs complex diseases. The unprecedented enrichment multi-omics data rapid development machine learning technology provide us opportunity design framework study relationship Results In this article, we proposed new approach, namely LGDLDA (LncRNA-Gene-Disease networks based LncRNA-Disease Association prediction), for disease-related prediction data, methods neural network neighborhood information aggregation. Firstly, calculates similarity matrix lncRNA, gene disease respectively, through lncRNA expression profile matrix, lncRNA-miRNA interaction lncRNA-protein matrix. We obtain by calculating lncRNA-gene gene-disease ontology, disease-miRNA Gaussian kernel similarity. Secondly, integrates matrices using nonlinear feature network. Thirdly, uses embedded node representations approximate observed matrices. Finally, ranks lncRNA-disease pairs then selects potential lncRNAs. Conclusions Compared methods, our method takes into account more critical obtains performance improvement cancer-related predictions. Randomly split experiment results show that stability better IDHI-MIRW, NCPLDA, LncDisAP NCPHLDA. on different simulation sets can accurately effectively predict Furthermore, applied three real including gastric cancer, colorectal breast