Multiple Data Integration Using Joint Non-negative Matrix Factorization

Manna, Sweta Roy, Indrani Majumder, Debapriya Banerjee, Ayan Pati, Soumen KumarNowadays, large-scale genomic studies make it possible to accumulate data on the same tumor tissue from multiple data sources. An in-depth study of multiomics data integration on tumor progression will contribute a lot in predicting medicine and detecting important biomarkers. Here, a novel method is proposed to integrate various biological data sources for identifying the abnormal functionality of gene module at the time of tumor progression. Integration of three different sources of biological dataset, genomic expression (GE), protein-protein interaction (PPI), and gene ontology (GO), through joint Non-Negative Matrix Factorization (jNMF) produced a single genomic meta-module. Meta-modules produced by jNMF that inherit the information of these data (GE, PPI, and GO) and contribute to identify the changes in modular structure between tumor and normal stage. The gene similarity of important GO terms is obtained from the PPI network of the meta-modules are analyzed for disease diagnosis. Finally, the changes in the interaction pattern of the dissimilarity between tumor and normal cell are identified by proposed method.