As recently stated, "GWAS have so far identified only a small fraction of the heritability of common diseases, so the ability to make meaningful predictions is still quite limited" (Collins, 2010). This "missing heritability" has been attribute to a number of potential causes, and it has become clear that most complex traits are influenced by many genes, each with effects too small to be reliably discovered using traditional analyses of GWAS data. We propose to develop several innovative approaches to enhance gene discovery and improve replication rates and generalization performance of predictive models. These methods will vastly increase the power to detect true (non-null) effects in data derived from current GWAS. While we emphasize applications to currently existing GWAS data for Inflammatory Bowel Disease and Cardiovascular Disease Risk Factors, the same methodological framework will be applicable to next generation sequencing data.
The Specific Aims of the proposal are:
Aim 1 : To Develop Statistical Methods Incorporating Functional Annotations that Improve Discovery Rates. We will develop and implement methods that extend current state-of-the-field analyses for GWAS of univariate phenotypes, using the LD-weighted SNP annotation methodology recently developed by our group. Specifically, we propose to extend the mixture model approach to account for SNP LD-weighted functional annotations.
Aim 2 : To Develop Statistical Methods Incorporating Pleiotropic Relationships that Improve Discovery Rates. We will generalize the mixture model approach to encompass covariance between z-scores of SNPs from two phenotypes simultaneously (i.e., pleiotropy) and to use the uncovered pleipotropic relationships to improve power for SNP discovery and replication.
Aim 3 : To Use Estimates from Empirical Bayes Models as Priors in Functional Characterization and Pathway Analyses. We will use posterior effect size estimates from pleiotropic Empirical Bayes analyses as inputs to explicate shared and unique genetic mechanisms of phenotypes, as well as molecular pathways.
Aim 4 : To Develop and Distribute Software. Computer software, implementing the methods developed in Aims 1-3, will be distributed as a freely available and user-friendly R package hosted on Bioconductor.org and as a suite of interactive GUI-based programs available on a website hosted by our lab.
We have recently demonstrated that by applying state-of-the-art statistical methods to large databases of GWAS summary statistics it is possible to greatly improve understanding of the underlying genetic basis of complex traits and disorders. Based on these published analyses, we propose to develop several innovative statistical approaches to enhance gene discovery and improve replication rates and generalization performance of predictive models;we will use these methods as inputs to improve functional characterization and pathway analyses of tag SNPs. The expected outcome of this study will be a set of novel statistical and computational approaches for improved discovery of genetic influences on human traits and diseases that can be applied to any phenotype represented in existing GWAS, or in deep sequencing studies.