TY - JOUR TI - Post GWAS analysis DO - https://doi.org/doi:10.7282/T3PV6N4V PY - 2015 AB - Genome-wide association studies (GWAS) have gained popularity in the past few years. Researchers have made findings on identifying genetic variants as risk factors for biological traits. A further question is how to apply the results to real-life applications. In this dissertation, a framework is proposed to conduct a post stage GWAS analysis, connect GWAS findings with existing clinical trials, and provide useful information to doctors and practitioners to help patients. The key part of this framework is to incorporate GWAS results with clinical information and perform appropriate analysis with the combined data. We illustrate the application to the Trial of Non-pharmacologic Interventions in the Elderly (TONE). TONE is a clinical trial for elderly with high blood pressure, in which patients were randomized to receive intensive intervention in weight if they were obese, sodium intake reduction, both weight and sodium control, or placebo. We investigate the relationship of 21 polymorphisms, which are reported to have association with hypertension, diabetes or obesity, with the change in systolic blood pressure at the end of the trial. The objective is to find the people who would significantly benefit from such interventions. For the analysis of data, we propose two approaches under the Post GWAS framework: recursive partitioning tree and exhaustive search. The recursive partitioning algorithm is a binary tree based algorithm that assigns different functionalities to SNP data and clinical data in the tree construction. We fit the tree to the data and examine the sensitivity of blood pressure drop given weight loss or sodium reduction. We compare classical regression tree with our modification to emphasize the differences in their structures. Tree methods are easy to interpret and compute, but only investigate a subset of the feature spaces. Exhaustive search is proposed to overcome this disadvantage. We look at all possible combinations of genotypes with sufficient sample and compute the sensitivities. We control multiplicity by the permutation version of false discovery rate method. Multidimensional scaling is used to determine the maximum number of polymorphisms to consider. Finally, we report and interpret the results from recursive tree and exhaustive search models. KW - Statistics and Biostatistics KW - Gene mapping KW - Genomics KW - Genomes--Analysis KW - Blood pressure LA - eng ER -