Abstract 5324: Pan-cancer identification of mutated pathways and protein complexes

Abstract

Large-scale cancer sequencing efforts such as The Cancer Genome Atlas (TCGA) and others have shown that tumors exhibit extensive mutational heterogeneity with relatively few genes mutated at significant frequency and many genes mutated in only a small number of individuals. This long tail phenomenon complicates the identification of driver mutations by their observed frequency. The long tail is explained in part by the fact that driver mutations target genes in signaling and regulatory pathways, and these pathways may be perturbed by different mutations in different tumors. We developed two complementary algorithms, HotNet2 and Dendrix++, to analyze combinations of mutations in known or novel pathways. HotNet2 uses prior knowledge of pathways and protein complexes represented in a genome-scale protein-protein interaction network, and identifies significantly mutated subnetworks using a heat diffusion model. HotNet2 simultaneously assesses both the significance of mutations in individual proteins and the local topology of protein interactions. Dendrix++ identifies combinations of mutations de novo, without prior knowledge of pathways or protein interactions, by finding sets of mutations that are mutually exclusive across the tumor cohort. There are numerous examples of mutually exclusive mutations between interacting proteins; e.g. BRAF and KRAS in colorectal cancer. Dendrix++ generalizes this idea to find larger groups of mutually exclusive mutations. We applied HotNet2 and Dendrix++ to whole-exome sequencing and copy number aberration data from 3299 samples of twelve tumor types from TCGA Pan-Cancer project. Both algorithms identified gene sets that overlap well-known cancer pathways (e.g. TP53, MAPK, and RAS signaling pathways), as well as genes and complexes with less characterized roles in cancer (e.g. the cohesin and condensin complexes). HotNet2 subnetworks also contained novel candidate cancer genes that were rarely mutated in this cohort and thus not reported by single-gene tests of significance, including KDM5A, SHPRH, and ARID4A, each of which interacts with well-known cancer genes. Many of these gene sets have biological functions often perturbed in cancer, such as chromatin modification (e.g. the SWI/SNF and BAP1 complexes) and DNA damage repair (e.g. SHPRH). These results demonstrate the ability of HotNet2 and Dendrix++ to identify novel combinations of mutations in thousands of tumors, prioritizing genes and mutations in the long tail for further experimental studies. Citation Format: Mark D. Leiserson, Fabio Vandin, Hsin-Ta Wu, Jason R. Dobson, Benjamin R. Raphael. Pan-cancer identification of mutated pathways and protein complexes. [abstract]. In: Proceedings of the 105th Annual Meeting of the American Association for Cancer Research; 2014 Apr 5-9; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2014;74(19 Suppl):Abstract nr 5324. doi:10.1158/1538-7445.AM2014-5324

DGPathinter: a novel model for identifying driver genes via knowledge-driven matrix factorization with prior knowledge from interactome and pathways

“…To evaluate the identification performance of DGPathinter, we compare our model with three existing methods, HotNet2 (Leiserson et al, 2014), NBS (Hofree et al, 2013) and MUFFINN (Cho et al, 2016), on datasets of three types of cancers: BRCA, GBM and THCA. In the performance evaluation, HotNet2, NBS and MUFFINN are set to their default parameters.…”

“…Nevertheless, although some driver genes are mutated at high frequencies among tumor samples, previous studies have reported that some driver genes are mutated at low frequencies, and the mutation frequencies of these genes are too low to be tested as statistically significant (Vandin, Upfal & Raphael, 2011;Leiserson et al, 2014;Raphael et al, 2014). A prevalent assumption to explain the long tail phenomenon is that genes usually interact with other genes, and some genes with no mutation can be perturbed by their interacting neighbors (Vandin, Upfal & Raphael, 2011;Leiserson et al, 2014;Raphael et al, 2014;Cho et al, 2016).…”

“…A prevalent assumption to explain the long tail phenomenon is that genes usually interact with other genes, and some genes with no mutation can be perturbed by their interacting neighbors (Vandin, Upfal & Raphael, 2011;Leiserson et al, 2014;Raphael et al, 2014;Cho et al, 2016). Based on this assumption, many studies for driver gene identification have been proposed by incorporating interactome information as prior knowledge (Vandin, Upfal & Raphael, 2011;Leiserson et al, 2014;Raphael et al, 2014;Hofree et al, 2013;Bashashati et al, 2012;Cho et al, 2016). The interactome information is employed as gene interaction network obtained from databases including iRefIndex (Razick, Magklaras & Donaldson, 2008), STRING (Szklarczyk et al, 2011) and others (Prasad et al, 2009;Lee et al, 2011;Das & Yu, 2012;Khurana et al, 2013).…”

“…The interactome information is employed as gene interaction network obtained from databases including iRefIndex (Razick, Magklaras & Donaldson, 2008), STRING (Szklarczyk et al, 2011) and others (Prasad et al, 2009;Lee et al, 2011;Das & Yu, 2012;Khurana et al, 2013). For example, HotNet and HotNet2 use the idea of heat-diffusion and propagate the mutation frequency scores of genes through the network, and calculate the significance scores of genes to identify potential driver genes (Vandin, Upfal & Raphael, 2011;Leiserson et al, 2014). NBS is an integrated method that propagates the mutations through the interaction network for each tumor sample as preprocessing, and uses matrix factorization to obtain mutation-based subtypes and the mutation profiles of each subtypes (Hofree et al, 2013), where the mutation profiles can be utilized to prioritize driver genes (Hofree et al, 2013;Shi, Gao & Wang, 2016).…”

“…In addition, interaction network information has also been used to predict patient specific driver genes, which helps the personalized analysis (Hou & Ma, 2014;Jia & Zhao, 2014;Bertrand et al, 2015). Through the network-based approaches, many novel potential driver genes have been discovered, which greatly complements the understanding of cancer driver genes (Leiserson et al, 2014;Raphael et al, 2014;Cho et al, 2016).…”