AbstractsMathematics

Association studies test for genetic variation influencing disease risk. We explore here the application and development of statistics for binary traits on family data. There are two main areas of focus: the first on comparing existing single-variant tests, and the second on developing a gene-based test. In the first part, we carried out a comparative study by applying 42 family-based association test statistics on different family-based datasets, which are simulated under a variety of scenarios (varying levels of linkage disequilibrium; dominant, additive, and recessive disease models; a variety of family structures). We have compared the Type I error, power and robustness of all the statistics. The results show that, when testing the null hypothesis of no association and no linkage, among the statistics that have well-behaved Type I error, the More powerful Quasi-likelihood Score test has the highest power and high robustness. In the second part, motivated by a need for powerful gene-based association statistics on family-based data for binary traits, we have proposed a new test statistic, which is based on a mixed model framework, Laplace's method and a variance component score test. We have compared the Type I error rates and power of our new statistic and six existing statistics by simulating different scenarios (varying the number and effect size of risk and protective variants). Our proposed statistic shows well-behaved Type I error and high power in some scenarios. The insights gathered here may improve public health by providing information on how to effectively utilize association methods to detect genetic variants that are related to disease. Ultimately, they should help improve the understanding of disease etiology.