Nipmi: A network method based on interaction part mutual information to detect characteristic genes from integrated data on multi-cancers

Comprehensive analysis of integrated data on multi-cancers is important for understanding the biological mechanism of these cancers at the system level. Network methods give us new insight into simultaneously identifying the characteristic genes and pathways from multi-cancers. Nevertheless, when measuring the similarity quantification of genes, it is a challenge to choose suitable methods for network construction and analysis. Herein, the NIPMI method, based on Interaction Part Mutual Information (IPMI) measure for detecting characteristic genes from multi-cancers data, is proposed. First, Robust PCA was applied to select genes for network construction. Then, a network construction measure, IPMI, was proposed to effectively quantify the similarity between genes in the network, which is the highlight of NIPMI. Furthermore, we introduced a novel topological property, Topological Score, that combined the local and global properties of each node to find more candidate nodes in the network. Finally, pathway enrichment analysis was performed to validate the biological functions of multi-cancers. The experimental results demonstrated that NIPMI facilitates the identification of characteristic genes in a multicancer network; thus, it may serve as a valuable tool for detecting characteristic genes and significantly enriched pathway terms.