|
Abstract:
|
With advancements in high-throughput technologies and reduced costs, a massive amount of genetic data has been collected in multiple studies, providing a great opportunity for meta-analysis of summary information from individual studies. Conventional methods has been commonly used in meta-analysis of information from single locus. Besides the single-locus analysis, researchers are often interested in studying a set of variants (e.g., variants in a gene). Therefore, there is a need for the development of new methods for multiple-variant meta-analysis. In this study, we propose a new kernel-based method for meta-analysis of multiple variants, considering their potential inter-relationships and their complex relationships with a phenotype. A stochastic gradient descent approach is also incorporated into the method, making it feasible to handle millions of samples. Through a preliminary simulation, we show that the method with stochastic gradient descent requires much less time and memory. Our simulation also shows that the new method outperforms an existing method when there is a complex relationship between variants and a phenotype.
|
