Research On Gene Association Detection Method Combining GWAS And EQTL

Mining genes that affect traits such as human diseases and crop yields has important research significance for treating complex diseases and improving crop quality.Genomewide association study(GWAS)provides new insights and directions for the study of many complex traits by identifying single nucleotide polymorphisms(Single Nucleotide Polymorphisms,SNPs).A large amount of GWAS summary data related to different complex traits has been accumulated.Studies have found that GWAS risk loci and expression quantitative trait loci(expression quantitative trait loci,e QTL)have a large number of overlaps in the non-coding regions of the genome.This phenomenon makes gene expression gradually become an important intermediary for revealing the regulatory role of GWAS.This paper studies the gene-trait association detection method that integrates GWAS and e QTL,mainly including the following two aspects:(1)Commonly used gene-trait association detection methods mainly include the following three types: gene-trait association detection method based on colocalization,gene-trait association detection method oriented to transcriptome-wide association study(TWAS)and gene-trait association detection method based on Mendelian randomization(MR).Some relevant reviews currently only introduce some of the three types of gene-trait association detection methods based on GWAS and e QTL,and there are no reviews that comprehensively and systematically introduce the three types of methods.In addition,the existing reviews did not conduct experiments to compare and analyze the performance and advantages and disadvantages of different algorithms.Aiming at the above problems,this paper first expounds three types of gene-trait association detection methods that integrate GWAS and e QTL,and compares and analyzes the differences,connections,advantages and disadvantages of different algorithms at the theoretical level.In order to further compare the performance of different methods,this paper summarizes the GWAS summary data sets of four blood lipid traits and the significant gene sets affecting four blood lipids reported in multiple studies.The above data sets were used to verify different methods,and the experiments were compared and analyzed in terms of method generalization performance,target gene detection results,and comparison with standard data sets.Finally,the advantages and limitations of various methods are analyzed from the perspective of experimental results.(2)The TWAS-oriented gene-trait association detection method is currently a commonly used gene-trait detection method,but it uses a traditional linear model when training the gene expression weight matrix,which cannot learn more nonlinear relationships between important mutation sites,which in turn affects the accuracy of the test results.To this end,this paper proposes a gene-trait association detection method based on Convolutional Neural Networks(CNN)and Transformer.Firstly,the convolutional neural network is used to extract the characteristics of important mutation sites,and then the Transformer is used to capture the dependencies between long-distance mutation sites,so as to improve the accuracy of gene expression prediction,and then improve the accuracy of gene-trait correlation detection.Based on the genotype and expression data of 339 different corn materials collected by the corn team of Huazhong Agricultural University,the experimental comparative analysis was carried out using phenotypic traits for flowering period.The results show that the gene-trait association detection method based on CNN and Transformer proposed in this paper is superior to the currently commonly used FUSION method in terms of the accuracy of gene expression prediction and the number of candidate genes.