Comparison of the performances of genomic relationship matrix calculation methods in genome-wide association studies

Studies conducted based on genome-wide marker data to detect sequence variants affecting various traits or diseases are called Genome-Wide Association Studies (GWAS). In the statistical models built for GWAS, the genomic relationship matrices (K) are used to avoid inflation in the number of false positives (FP) that may arise from genetic relatedness of samples. Various methods have been developed to calculate K from SNP marker data. In this study, the performances of the methods Van Raden 1 (VR1), Van Raden 2 (VR2), Astle-Balding (AB), Yang (UARA), Identity by State (IBS), Simple Matching with Correction (SMC) and Gaussian Kernel (GK) were compared for detecting True Positives (TPs) and reducing FPs in GWAS analysis with the FarmCPU method. For this purpose, GWAS analyzes were performed using the SNPs of 4495 individual goats from 32 different goat breeds, and three simulated traits from these SNPs with 0.1, 0.3 and 0.5 heritability. It was determined that there were no significant differences between the K matrix calculation methods in terms of TP detection performance, but AB, VR2 and UARA methods were relatively better in terms of reducing the number of FPs, especially for high heritability traits. Among the methods in question, the AB method can be preferred when working with large-size genotypic data due to its shorter calculation time. The findings revealed that the TP detection performance of GWAS models increases closely with the heritability of the traits. In addition, it has been determined that performing GWAS analysis with a PC number below the optimal PC number determined from the first bending point in the scree plot in the PCA analysis can be used as a rapid preliminary discovery option for TP detection before a well-configured GWAS analysis. Keywords:Genomic relationship matrices, Genom-wide association studies, Genomic data analysis, Statistical models, Animal breeding