Enhancing Biological Insights Through Statistical Inferences
Date of Award
Doctor of Philosophy (PhD)
In genetic association studies, correlation among genetic variants and among phenotypes makes interpretation of the test results difficult. When correlated genetic variants are present, correlated SNPs display similar association results making interpretation difficult. When correlated phenotypes are present, it is hard to discern whether an associated genetic variant shows effect on one phenotype or the other. In this thesis we completed two projects that share a common theme of trying to overcome these effects of correlation. In the first project we developed a new method that identifies conditions under which observed univariate and joint results for two correlated SNPs can be produced by a single causal SNP, D. This method identifies properties of D (minor allele frequency, pair-wise correlations to the two SNPs, and odds ratio) that would give rise to an observed joint SNP analysis result of two SNPs. Real SNPs that match an identified minor allele frequency and pair-wise correlations can be considered candidates for D. We learned that D may have only modest correlation to the two SNPs, and D can be a common SNP with moderate odds ratio. Therefore traditional approach of searching for causal variants among the "tag" that are highly correlated with the two SNPs would not have found D. We also demonstrate the utility of this method by applying it to our previously reported joint SNP results for nicotine dependence [Saccone, et al. 2009]. In this application, we do not find any D that can account for the association test result of rs16969968 and rs588765. This evidence further supports the interpretation that these two SNPs represent two distinct associations. In the second project, we conducted a collaborative meta-analysis of a non-synonymous variant rs1799971, an asparagine to aspartic acid change with respect to multiple substance dependences that are correlated to one another (alcohol, nicotine, marijuana, cocaine, and opioids). Our collaborative approach improves on literature-based meta-analyses by conducting new, coordinated analyses across multiple datasets, both published and unpublished. We report evidence that rs1799971 shows a modest protective effect on general substance dependence (cases are dependent on any substance). Furthermore, we observed that the point estimate odds ratio for the effect of rs1799971 on each individual substance was similar to that observed for general substance dependence, although no single substance dependence showed significant association. This finding indicates that rs1799971 may contribute to shared, non-substance-specific risk for addiction. Further, our finding shows the power of large collaborative efforts that can detect even modest associations and also points out the importance and the need of pooling resources to accelerate discoveries. Overall, the work in this thesis provides improved understanding of association results to provide a better set of biological targets for follow-up studies.
Chair and Committee
Nancy L Saccone
Laura Bierut, Anne Bowcock, Don Conrad, Michael Province, John Rice, Cristina Strong
An, Tae-Hwi, "Enhancing Biological Insights Through Statistical Inferences" (2013). Arts & Sciences Electronic Theses and Dissertations. 53.
Permanent URL: https://doi.org/10.7936/K7WW7FKX