Bayesian estimation of gene effects on traits from coding variants - Abstract One central goal of human genetics is to determine the key genes and molecular pathways that drive disease. By doing so, we can hope to gain deeper insight into the molecular basis of disease, as well as to identify potential targets for therapeutic intervention. While GWAS has identified tens of thousands of significant, robust associations, the results do not provide a straightforward assessment of the potential importance of each gene. As an alternative, we propose to develop new techniques for using rare protein-coding variants including loss- of-function mutations, deletions, and duplications in burden tests. We will use these to estimate the magnitude and direction of effect of each gene on a phenotype of interest, across the full range of expression encompassed by natural variation. In order to improve accuracy, since tests at single genes are often underpowered, we will extend a machine learning approach we developed previously called GeneBayes to share information among similar genes within a hierarchical Bayesian framework. We will release open access software and summary statistics to maximize the value of our work to the scientific community.
