Statistical methods to detect genotype-phenotype association using genetic similarity matrices
MetadataShow full item record
The main focus of this thesis is to investigate and compare different statistical methods to perform genotype-phenotype association analyses of HUNT VO2max data from 1472 men, while accounting for genetic confounding. The methods of interest are fitting a linear regression model to a reduced sample of 1274 men, and fitting a linear mixed model to the full sample, with maximal oxygen uptake as response, and age and activity index as covariates. The covariance matrix of the linear mixed model is a scaled version of an estimated genetic similarity matrix, the kinship matrix, estimated from 102 477 SNPs. Each SNP is then tested for association with the response using a score test. The analyses are performed using GenABEL, which is an R-package for statistical analyses of genome-wide association studies. We analyze only the 9069 SNPs on chromosome 1. The results from both methods show no significant associations between any SNP and the VO2max, when controlling the family-wise error rate at level 0.05. Based on the results of the most significant SNPs and estimation of the genomic control inflation factor for both methods, we find that the preferred procedure for performing genotype--phenotype association analyses is using linear mixed models. The incorporation of the estimated kinship matrix accounts for the correlation between the individuals caused by population structure and cryptic relatedness.