Pan-cancer analysis of somatic mutations and transcriptomes reveals common functional gene clusters shared by multiple cancer types

Hyeongmin Kim,Yong-Min Kim

doi:10.1038/s41598-018-24379-y

Hyeongmin Kim, Yong-Min Kim

Open Access

https://doi.org/10.1038/s41598-018-24379-y

Copy DOI

Abstract

To discover functional gene clusters across cancers, we performed a systematic pan-cancer analysis of 33 cancer types. We identified genes that were associated with somatic mutations and were the cores of a co-expression network. We found that multiple cancer types have relatively exclusive hub genes individually; however, the hub genes cooperate with each other based on their functional relationship. When we built a protein-protein interaction network of hub genes and found nine functional gene clusters across cancer types, the gene clusters divided not only the region of the network map, but also the function of the network by their distinct roles related to the development and progression of cancer. This functional relationship between the clusters and cancers was underpinned by the high expression of module genes and enrichment of programmed cell death, and known candidate cancer genes. In addition to protein-coding hub genes, non-coding hub genes had a possible relationship with cancer. Overall, our approach of investigating cancer genes enabled finding pan-cancer hub genes and common functional gene clusters shared by multiple cancer types based on the expression status of the primary tumour and the functional relationship of genes in the biological network.

Highlights

Protein-protein interaction (PPI) network of protein-coding Pan-cancer-wide selected genes (PSGs)
Besides protein-coding PSGs (pcPSGs), we found evidence that non-coding PSGs (ncPSGs) were related to cancer
Among ncPSGs, pseudogenes accounted for 77.2% (Supplementary Fig. S13)

Summary

Introduction

Protein-protein interaction (PPI) network of protein-coding PSGs (pcPSGs). For pcPSGs, we generated a single-depth PPI network and discovered subnetworks. We used information of PPI from STRING V1017 with stringent cut off that was a combined interaction score ≥ 900. To find gene clusters and create subnetworks, we used MCODE15 plugin of Cytoscape[52]. The program options ‘fluff ’ and ‘K-core = 10’ were used to increase the size of subnetworks and filter out clusters lacking a maximally interconnected node of at least 10 degrees of edges. Because the ‘fluff ’ option was used, genes of clusters were partially overlapp

Objectives

Methods

Results

Conclusion