GWAS & Selection signatures

A Novel Monte Carlo Method for Approximating the Null Hypothesis in Genomic-Windowà¢â‚¬â€œBased Genome-Wide Association Analyses

---

Abstract

Genome‐wide association studies (GWAS) are commonly used to identify QTL. Several authors have proposed using genomic windows rather than testing individual SNP significance to mitigate the limited information from single markers. However, GWAS results based on genomic windows can be strongly influenced by SNP density, linkage disequilibrium, allele frequencies, and genetic similarity among individuals within each region. In this study, we propose a new Monte Carlo approximation to generate a null hypothesis that may correct these biases. The core idea is to simulate populations sharing the characteristics of the observed dataset. Phenotypes in simulated populations were generated by adding systematic, random environmental, additive genetic, and residual effects. Environmental and residual effects were sampled from Gaussian distributions with corresponding variances. Additive genetic effects were sampled from a multivariate Gaussian distribution with mean zero and variance-covariance matrix Hσa2. Samples were obtained from the stationary distribution of an updating scheme based on conditional distributions of each additive effect computed through the H-1 matrix. To avoid autocorrelation, one set of breeding values was retained every 100 iterations. The procedure was tested using cold carcass weight data from Rubia Gallega cattle, including 99,227 records and a pedigree of 150,904 individuals. A total of 5,627 individuals were genotyped for 39,194 SNPs. Variance components were estimated with the AIREML algorithm in blupf90+, yielding additive, herd-year-season, and residual variances of 494.85, 290.86, and 456.85 kg2, respectively. The software postgsf90 identified seven 1-Mb genomic regions explaining >0.5% of additive variance, while the SNP-GWAS only identified two genomic regions with p-values of 7.16 x 10E-11 and 1.05 x 10E-6, near the well-known MSTN and LCORL genes. The null distribution was generated by simulating 1,000 populations that were analyzed with postgsf90. The null distribution was highly heterogeneous, with mean values ranging from 0 to 1.73% of additive variance. The p-values were calculated by assuming one gamma distribution for each genomic region. The results showed that 6 of the 7 original regions were false positives, while one of the genomic regions initially below 0.5% became significant. Moreover, the two regions previously detected with the SNP-GWAS achieved higher levels of significance, with p-values of 4.44 x 10E-15 and 7.13 x 10E-10, respectively. Additionally, another genomic region in the BTA19 was identified with a p-value of 5.8 x 10E-6. This study demonstrates that windows variance approach is affected by multiple biases, and that the proposed Monte Carlo procedure effectively corrects them. Moreover, it also confirms that the windows variance approach can be more powerful that the SNP-GWAS. Nevertheless, the results are conditioned on the variance component estimates. Further studies should be developed to generate the null hypothesis from the posterior distribution of the variance components within the Bayesian framework.

Keywords: 2026

How to Cite:

López-Carbonell, D., Sánchez-Díaz, M., Barceló-Blasco, F. & Varona, L., (2026) “A Novel Monte Carlo Method for Approximating the Null Hypothesis in Genomic-Windowà¢â‚¬â€œBased Genome-Wide Association Analyses”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2285397. doi: https://doi.org/10.31274/wcgalp.23682

Rights: 1