A Statistic Model Of Identifying Mutual Exclusivity Mutations In Cancer Pathway

Objective:Advances in high-throughput sequencing technologies and decreases in the cost of sequencing have enabled TCGA and ICGC collecting amounts of cancer genome mutation data. Identification and prediction of cancer genes have become key issues in bioinformatics research. Our study is focused on the research of the "mutually exclusive mutations" identification methods, especially for the basic principles of MEGSA algorithm. We first evaluate the performance of it in simulation study, and then apply it to GBM data which contains SNP and CNV to identify the optimal mutually exclusive gene sets, and provide evidence for gene drug development and cancer diagnosis or treatment. Methods:We summed up the ideal analytic framework for identifying MEGS through the study of the existing methods for identifying MEGS. MEGSA algorithm is now the optimal algorithm to identify MEGS. During the simulation analysis, we seted MEGS = 3, Coverage = 0.1,0.2,0.25,0.3,0.4,0.5; sample size = 50,100,200,300,400,500; the number of randomly mutated genes= 10,15,20,25,30; the random mutation rate = 10% to construct mutation matrix, then compared the power of detecting true MEGS in different parameter settings. According to the GBM data(261 patients and 398 alterations), we first organized it into binary mutation matrix, and then analyzed it with the R software. Finally we initially found the mutually exclusive gene sets in GBM pathways。 Results:The simulation result showed that the power of MEGSA for identifying MEGS increasing with the increases of coverage and sample size. For GBM, we identified 20 significant MEGS which contains 12 gene variants, SNP: RB1 mutation, TP53 mutation, IDH1 mutation, PTEN mutation, NF1 mutation, SPTA1 mutation; CNV: CDK4 amplification, CDKN2 A deletion, MDM2 amplification, EGFR amplification, PTEN deletion, PDGFRA deletion and a CNV meta genes(MET, CAP2A2, ST7-AS1, ST7, ST7-OT4). The most significant MEGS contains three genes: CDK4 amplification, CDKN2 A deletion and RB1 mutation. Conclusion:MEGSA is a relatively flexible and powerful method. It performs well not only for point mutation data, but also for the data contained copy number variations. Compared with Multi-Dendrix algorithm, MEGSA had a higher power. The identified mutations had been proved in GBM pathogenic pathways except for SPTA1.The identified CNV also confirmed related to the development of cancer. But MEGSA used multi-path search algorithm for calculating, some results may be lost in this way. And permutation test requires more times to get the desired result, this will need greater computer CPU and memory.