Network-Based Methods For Mining Cancer Related Driver Patterns

Decades of research have shown that cancer is a complex disease caused by mutations in the genome.With the development of science and technology,the cost of genome sequencing has been rapidly reduced,and the massive data of cancer genomics has provided great support of further study on cancer pathogenesis.Currently,one of the key challenges in cancer genomics is to detect mutated genes that are involved in the development and progression of cancer from somatic mutation data for cancer,known as driver genes.However,gene mutations have very high heterogeneity in a large number of cancer patients,and the research methods for identifying driver mutations or driver genes based on background mutation rate cannot effectively solve this problem.Many studies have shown that different driver genes can perturb the same biological pathways,therefore detecting the cancer-related biological pathways can take mutation heterogeneity into account and further study the carcinogenic mechanism at pathway level.Many biologists observe that driver genes exhibit two properties at the pathway level: high mutual exclusivity and high coverage.And a lot of studies have proposed many methods to identify driver pathways or modules based on these two properties.In this dissertation,based on the shortcomings of the methods to detect cancer driver pathways or modules,we propose the network based methods for mining different types of cancer driver pathways or modules.The driver module types in this dissertation include driver modules in a single cancer,driver modules with rarely mutated genes in multiple cancer types or single cancer type,specific driver modules for a certain cancer type in the context of multiple cancer types,and common driver modules for multiple cancer types.These different types of cancer driver modules offer support for understanding the molecular mechanisms of cancer from different perspectives.The main research work and innovations are as follows: 1.For the problem of detecting driver modules,this dissertation proposes an index to measure the mutual exclusivity between two mutated genes,which can more reasonably capture the weak mutual exclusivity between two genes.Then,by combining the mutual exclusivity and coverage,a mutated gene network is constructed,the fully connected subgraphs are detected from the network,and the cancer driver modules are detected through a strict screening strategy.This method is applied to the somatic mutation data sets of glioblastoma and breast cancer,and it can detect the cancer related driver modules successfully.When compared with other methods on pathway enrichment,this method has higher accuracy.2.Rarely mutated genes play key roles in the development and progression of cancer.However,few methods are developed to study the carcinogenic mechanism of rarely mutated genes at the pathway level.In this dissertation,a functional similarity index is introduced to measure the functional relationship between rarely mutated genes and other genes in the same pathway.In combination with mutual exclusivity and coverage,a functional mutated gene network is constructed,and driver modules with rarely mutated genes are detected from the network.This method successfully detects the cancer related driver modules with rarely mutated genes in multiple cancer types and single cancer types.And the rarely mutated genes in the detected driver modules are always known to be cancer genes.When compared with other methods,this method can identify more rarely mutated genes and has higher pathway enrichment.By discussing the importance of the functional similarity,mutual exclusivity and coverage index,it shows that the functional similarity index tends to be the dominant factor.3.The methods of identifying driver modules are generally designed for a single cancer type,but few methods consider the differences and similarities among multiple cancers.This dissertation presents a network based framework for detecting cancer specific and common driver modules.The framework for the identification of cancer specific driver modules(CSDM)considers the specific coverage and mutual exclusivity of a certain cancer type,builds a specific network of this cancer to other cancers,and uses the greedy algorithm to detect the specific cancer driver modules in its specific network.At the same time,the framework for identifying cancer common driver modules(CCDM)constructs a common network for multiple cancers according to the coverage stability and mutual exclusivity stability of each gene pairs.Then,a greedy algorithm is used to detect common driver modules in the common network.CSDM successfully identifies specific driver modules for a cancer to other cancers,and it has higher module specific coverage and pathway enrichment than other methods.CCDM is effective in detecting common driver modules in multiple cancer types,and it has higher module coverage stability and pathway enrichment than other methods.