Prior knowledge driven joint NMF algorithm for ceRNA co-module identification

MRNA and lncRNA serve as a type of endogenous RNA in cell, which can competitively bind to the same miRNA through miRNA response elements (MREs), thereby regulating their respective expression levels, playing an important role in post-transcriptional regulation, and regulating the progress of tumors. The proposed competing endogenous RNA (ceRNA) hypothesis provides novel clues for the occurrence and development of tumors, but the integrative analysis methods of diverse RNA data are significantly limited. In order to find out the relationship among miRNA, mRNA and lncRNA, the previous studies only used individual dataset as seeds to search two other related data in the database to construct ceRNA network, but it was difficult to identify the synchronized effects from multiple regulatory levels. Here, we developed the joint matrix factorization method integrating prior knowledge to map the three types of RNA data of lung cancer to the common coordinate system and construct the ceRNA network corresponding to the common module. The results show that more than 90% of the modules are closely related to cancer, including lung cancer. Furthermore, the resulting ceRNA network not only accurately excavates the known correlation of the three types of RNA molecular, but also further discovers the potential biological associations of them. Our work provides support and foundation for future biological validation how competitive relationships of multiple RNAs affects the development of tumors.