Study Of Cancer Driver Mutation Discovery Algorithms Oriented Toward Tumor Heterogeneity

Cancer is mainly caused by genomic mutations.Discovering cancer driver mutations which are contributing to cancer from cancer genomic alterations is important for the research of cancer diagnosis and treatment.With the development of high-throughput microarray technology and next-generation sequencing technologies,the existing studies have accumulated a large amount of genomic mutation data of cancer patients,which provides the opportunity to discover cancer driven mutations computationally.In the existing studies,the drive mutation discovery methods mainly detect the occurring frequencies of mutation in cancer samples,and select mutations with high frequency as a cancer driven mutations.However,recent studies show that cancers exhibit tumor heterogeneity,where different mutations occur in different samples of cancers.In heterogenous cancers,some driver mutations only occur in a part of samples,and these mutations display relatively low frequencies in all samples.Therefore,it is difficult for the existing methods to effectively detect the driver mutations aforementioned.To effectively detect driver mutations from heterogeneous cancers,we improve the matrix factorization algorithm according to the features of cancer genomic mutations,and propose cancer driver mutation discovery algorithms oriented toward tumor heterogeneity.The main contritbutions of this dissertation are listed below:1.To discover driver mutation from copy number aberrations,the existing methods mainly focus on detecting recurrent copy number aberrations.In this study,we modify the sparse singular value decomposition algorithm,and esitablish an algorithm to discover recurrent copy number aberrations with complex patterns.Based on the aforemented research,we further propose an algorithm to discover subgroup-specific recurrent copy number aberrations for heterogeneous cancers.The evaluation results illustrate that the proposed algorithms have great advantage in driver mutation discovery of heterogeneous cancers.2.For single nucleotide variation data of heterogeneous cancers,we propose a novel network regularized matrix factorization algorithm to discover driver mutation related genes.This algorithm can effectively measure the mutation frequencies of genes in a part of samples,and incorporate the information of gene interaction network by using graph Laplacian regularization.When we analyze the detection results of the algorithm on various types of cancers,the evaluation results indicate that the driver mutation related genes discovery performance of the proposed algorithm is better than the existing methods.3.In pan-cancer analysis where the combination of multiple types of cancers is used,by considering the-tumor heterogeneity of pan-cancer data and the relationship between different types of cancers,we propose a driver mutation related gene discovery algorithm oriented toward pan-cancer analysis,which is based on matrix tri-factorization.Meanwhile,we also introduces similarity information between different types of cancer by using pairwise similarity constraint.The results show that the proposed algorithm can effectively identify driver mutation related genes from pan-cancer data.