| Genome-wide Association Studies(GWAS)refers to the association analysis of genotypes and phenotypic traits/diseases on the whole genome scale.Single nucleotide polymorphism(SNP)is a kind of genetic molecular marker,which is numerous in genome and easy to detect.GWAS usually uses SNP as a marker to study the pathogenic genes of complex diseases,and the study of genome-wide SNP interaction has become an important method to explore the pathogenic mechanism of complex diseases.Due to the large-scale genome-wide datasets,heavy computational burden and diversity of disease models,some existing methods still have some drawbacks,such as low detection power,high false positive rate,slow running speed and bias of disease models.Therefore,it is very important to study to design the accuracy and fast SNP interaction detection methods for finding the pathogenic genes of complex diseases.To solve the problems of low detection efficiency,high false positive rate and biased disease model,a SNP interaction recognition algorithm DSNPE-IMOCS,based on multi-objective optimization cuckoo search algorithm,was proposed.Bayesian network and Gini coefficient are used as two objective functions to fit different disease models,uses fast non dominated sorting,crowding distance and Pareto strategy to select the nondominated frontier solution set of SNP combination solution set,and the optimal non dominated frontier solution set of SNP combination that may be associated with disease can be obtained by cuckoo search algorithm,such improve the detection efficiency;The candidate filtering strategy is used to filter non-dominant frontier solutions,so that reduce the false positive rate of the candidate solutions;and the G-test statistical test method is used to test all SNP combinations to obtain the truly pathogenic SNP interaction combination,the corresponding p-value and crowding distance value.The experimental results on DME(Displaying SNP Interaction and Marginal effects)and DNME(Displaying SNP Interaction with No Marginal Effects)simulation data set,and real data set show that DSNPE-IMOCS algorithm has higher detection power and precision,compared with similar SNP interaction detection methods,and reduces false positive rate.To accelerate the multi-objective optimization of genome-wide SNP interaction detection,a GPU parallel computing detection algorithm DSNPE-IMOCS-GPU is designed.The algorithm parallelizes the most time-consuming multi-objective optimization cuckoo search stage in DSNPE-IMOCS algorithm.In this stage,each operation that can be parallelized is mapped to a kernel function that meets CUDA parallel programming model,making the time-consuming search running on GPU,while the simple candidate solution filtering and G-test verification work are designed to be executed on CPU,thus avoiding frequent data transferring between CPU and GPU to achieve accelerated calculation and reduce the running time of the algorithm.The experimental results of the DNME simulation data set and real data set show that the GPU parallel detection algorithm DSNPE-IMOCS-GPU achieves the same detection power as the DSNPE-IMOCS algorithm generally.At the same time,it can successfully detect SNPs loci related to Age-related Macular Degeneration(AMD),and the SNP interaction combination(rs6967345,rs3775640)detected by DSNPE-IMOCS-GPU for the first time,which may be related to AMD.DSNPE-IMOCS-GPU also reduces the average run time required for the detection process and accelerates the effect significantly. |
