Research And Applications Of Statistical Methods For Genetic Variants Association Analysis

Many diseases of human are complex diseases whose occurrence and development processes may be influenced by genetic and environmental factors. So, searching for causal genetic locus is important for human being to understand the pathogenesis of complex diseases, and to find the methods of correct diagnosis, treatment and prevention.Genome-wide association studies have successfully identified thousands of common genetic variants associated with human traits or complex diseases. As the development of study, we have found that many complex diseases were associated with rare variants.Therefore, the association analysis of genetic variants has become hot topic in recent years. The associations between disease and genetic variants are investigated by some statistical methods in this dissertation.Firstly, the association between genetic variants and binary trait is analyzed by the combination of P-values. Each rare variant is tested by Fisher’s exact test, and each common variant is tested by the score test, then per-site P-values are obtained. All genetic variants are divided into deleterious variants and protective variants according to the number of minor allele in cases and controls. According to deleterious variants and protective variants, per-site P-values are separately combined with a suited weight scheme.The weights of rare variants are beta distribution density functions with parameters 1 and25 respectively, the weights of common variants are beta distribution density functions with parameters 0.5 and 0.5 respectively. To guard against the noise caused by noncausal variants, variants with P-values larger than a threshold will be truncated. The threshold is not fixed, on the contrary, multiple candidate thresholds are allowed, then the optimal threshold is chosen for any given data set.Secondly, the association between rare variants and quantitative trait is analyzed by the combination of P-values. Each rare variant in a region is tested by the score test, covariates are simultaneously adjusted, and then per-site P-values are obtained. All genetic variants are divided into deleterious variants and protective variants according to the mean value of trait values of individuals with minor allele and without minor allele. According to deleterious variants and protective variants, per-site P-values are separately combined with weight. The weights of variants are beta distribution density functions with parameters 1 and 25 respectively. Our proposed two methods are robust to the different directions of effects of causal variants and the inclusion of a large proportion of noncausal variants.Finally, the association between rare variants and quantitative trait is analyzed under the design of extreme phenotype sampling. Considering the sample design firstly, sampling individuals with extreme phenotypes can enrich the frequency of rare variants. The extreme phenotype sampling is regarded as binary trait, the individuals with higher trait value as cases and the individuals with lower trait value as controls. A logistic model is used for these “case-control”data, the effects of all genetic variants are divided into the common effect and the individual effects deviation from the common effect. The common effect is regarded as fixed effect, the individual effects are regarded as random effects. Two classes effects are jointly tested, and then separate P-values are obtained.The test statistics are obtained by Fisher’s method and minimum-P approach of combining of P-values. A large of simulation studies show that our proposed methods are more powerful in various settings of models, sampling from extreme phenotypes outperforms random sampling methods when the same sample size is used.