Recurrence is a hallmark of cancer-driving mutations. Recurrent mutations can arise at the same site or affect the same gene at different sites. Here we identified a set of mutations arising in individual samples and altering different cis-regulatory elements that converge on a common gene via chromatin interactions. The mutations and genes identified in this fashion showed strong relevance to cancer, in contrast to noncoding mutations with site-specific recurrence only. We developed a prediction method that identifies potentially recurrent mutations on the basis of the features shared by mutations whose recurrence is observed in a given cohort. Our method was capable of accurately predicting recurrent mutations at the level of target genes but not mutations recurring at the same site. We experimentally validated predicted mutations in distal regulatory regions of the TERT gene. In conclusion, we propose a novel approach to discovering potential cancer-driving mutations in noncoding regions.

BackgroundOne of the greatest challenges in cancer genomics is to distinguish driver mutations from passenger mutations. Whereas recurrence is a hallmark of driver mutations, it is difficult to observe recurring noncoding mutations owing to a limited amount of whole-genome sequenced samples. Hence, it is required to develop a method to predict potentially recurrent mutations.ResultsIn this work, we developed a random forest classifier that predicts regulatory mutations that may recur based on the features of the mutations repeatedly appearing in a given cohort. With breast cancer as a model, we profiled 35 quantitative features describing genetic and epigenetic signals at the mutation site, transcription factors whose binding motif was disrupted by the mutation, and genes targeted by long-range chromatin interactions. A true set of mutations for machine learning was generated by interrogating publicly available pan-cancer genomes based on our statistical model of mutation recurrence. The performance of our random forest classifier was evaluated by cross validations. The variable importance of each feature in the classification of mutations was investigated. Our statistical recurrence model for the random forest classifier showed an area under the curve (AUC) of ~0.78 in predicting recurrent mutations. Chromatin accessibility at the mutation sites, the distance from the mutations to known cancer risk loci, and the role of the target genes in the regulatory or protein interaction network were among the most important variables.ConclusionsOur methods enable to characterize recurrent regulatory mutations using a limited number of whole-genome samples, and based on the characterization, to predict potential driver mutations whose recurrence is not found in the given samples but likely to be observed with additional samples.Electronic supplementary materialThe online version of this article (doi:10.1186/s12859-016-1385-y) contains supplementary material, which is available to authorized users.

Global network modeling of distal regulatory interactions is essential in understanding the overall architecture of gene expression programs. Here, we developed a Bayesian probabilistic model and computational method for global causal network construction with breast cancer as a model. Whereas physical regulator binding was well supported by gene expression causality in general, distal elements in intragenic regions or loci distant from the target gene exhibited particularly strong functional effects. Modeling the action of long-range enhancers was critical in recovering true biological interactions with increased coverage and specificity overall and unraveling regulatory complexity underlying tumor subclasses and drug responses in particular. Transcriptional cancer drivers and risk genes were discovered based on the network analysis of somatic and genetic cancer-related DNA variants. Notably, we observed that the risk genes were functionally downstream of the cancer drivers and were selectively susceptible to network perturbation by tumorigenic changes in their upstream drivers. Furthermore, cancer risk alleles tended to increase the susceptibility of the transcription of their associated genes. These findings suggest that transcriptional cancer drivers selectively induce a combinatorial misregulation of downstream risk genes, and that genetic risk factors, mostly residing in distal regulatory regions, increase transcriptional susceptibility to upstream cancer-driving somatic changes.

Major progress in disease genetics has been made through genome-wide association studies (GWASs). One of the key tasks for post-GWAS analyses is to identify causal noncoding variants with regulatory function. Here, on the basis of >2000 functional features, we developed a convolutional neural network framework for combinatorial, nonlinear modeling of complex patterns shared by risk variants scattered among multiple associated loci. When applied for major psychiatric disorders and autoimmune diseases, neural and immune features, respectively, exhibited high explanatory power while reflecting the pathophysiology of the relevant disease. The predicted causal variants were concentrated in active regulatory regions of relevant cell types and tended to be in physical contact with transcription factors while residing in evolutionarily conserved regions and resulting in expression changes of genes related to the given disease. We demonstrate some examples of novel candidate causal variants and associated genes. Our method is expected to contribute to the identification and functional interpretation of potential causal noncoding variants in post-GWAS analyses.

Major progress in disease genetics has been made through genome-wide association studies (GWASs). One of the key tasks for post-GWAS analyses is to identify causal noncoding variants with regulatory function. Here, on the basis of > 2,000 functional features, we developed a convolutional neural network framework for combinatorial, nonlinear modeling of complex patterns shared by risk variants scattered among multiple associated loci. When applied for major psychiatric disorders and autoimmune diseases, neural and immune features, respectively, exhibited high explanatory power while reflecting the pathophysiology of the relevant disease. The predicted causal variants were concentrated in active regulatory regions of relevant cell types and tended to be in physical contact with transcription factors while residing in evolutionarily conserved regions and resulting in expression changes of genes related to the given disease. We demonstrate some examples of novel candidate causal variants and associated genes. Our method is expected to contribute to the identification and functional interpretation of causal noncoding variants in post-GWAS analyses.