Mutated Tumor Suppressors Follow Oncogenes Profile by the Gene Hypermethylation of Partners in the Protein Interaction Networks

Abstract

SummaryAs a result of current advances in the analysis of patient sequencing data, many tumors have been characterized in a personalized manner. Such data can also be used to characterize genes that act as either oncogenes or tumor suppressors. These include “defective” tumor suppressor genes which may function as driver oncogenes that play a key role in cancer proliferation due to various genetic alterations, specifically, chromosomal translocations. In this study, we considered protein networks, mutations, methylation data and cancer fusions to classify tumor suppressors that may convert into oncogenes. Moreover, we developed a novel network-based parameter called the ‘preferential attachment score’ to categorize genes as oncogenes and/or tumor suppressors. Such classification was achieved using a naïve Bayes computation approach. We used an ABC-MCMC method for selecting features for training our classification algorithm. We then performed a survey of tumor suppressors and oncogenes from the perspective of somatic mutations and network properties for 691 TCGA cases. For comparative purposes, we chose currently well-established methods, such as MutSigCV, OncodriveCLUST, Oncodrive-FM, 20/20+, ActiveDriver, MuSiC, TUSON, OncodriveFML, and found that our algorithm outperformed these other tolls, with 93.3% efficiency. Based on 691 TCGA cohorts, we found that tumor suppressors presented the highest mutation frequency in most tumor types, relative to oncogenes. Using protein-protein interaction data, we found that essential proteins, tumor suppressors and oncogenes had higher degrees of connectivity and betweenness centrality, relative to normal proteins. Similarly, tumor suppressors and oncogenes had lower clustering coefficients, as well as shortest path distances (FDR < 0.05). Finally, most mutated tumor suppressors integrate hyper-methylated partners in the protein interaction networks of 3091 fusions, following the patterns of oncogenes (43%). Thus, these results further characterize cancer oncogenes and tumor suppressors in the context of deep analysis of cancer network alterations.AvailabilitySource scripts are available at https://github.com/somnathtagore/NBC and the resource is available at http://ontum.md.biu.ac.il/index.html