Statistical Analysis Methods for Genetic and Epigenetic Data

点击翻译按钮获取中文摘要

Despite the thousands of genome-wide association studies (GWAS) that have been published in the past 15 years there are still major issues in the field. Many findings have failed to replicate in independent studies suggesting false positives. On the other hand, taking all of the markers found so far for any given trait typically only explains a small fraction of the estimated heritability of that trait suggesting many false negatives. Part of my research program is focused on addressing this lack of power in genetic studies. I plan to do this in two ways: 1. increase power to detect real effects by focusing on markers that are more likely to show an effect on a trait and 2. reduce the multiple testing burden and increase the total effect size by examining multiple markers at once rather than each one individually. To date most statistical testing in GWAS ignores external biological knowledge in an attempt to avoid bias. Years of genetic and genomic work, however, means that we now have very reliable annotations on the function of many genetic markers. In recent years a number of researchers have developed scores that rank variants in terms of their likelihood to have a biological effect. Usually, however, these are only used after a GWAS to decide on which top hits the researchers should focus. I propose to introduce this information earlier and use it to help inform our analysis of the genotype data. I will develop methods that will focus attention primarily on markers that are more likely apriori to have effects. In separate work I will also examine better methods to combine data from multiple markers so that association at the level of a gene or genomic region can be assessed. Doing this will increase biological relevance and also improve power by reducing the multiple testing burden and increasing the estimated effect sizes. Other reasons for the “missing heritability” is that not all heritable effects come through DNA alone. DNA methylation is heritable but does not change the underlying genetic code. For methylation analysis to be successful we must carefully normalize the data to remove non-biological variability. In this program I will examine existing ways that this is done and develop improved methods. I will also examine the question of association of phenotypic traits with a particular form of methylation which is abundant in the human brain and of great interest to neurological researchers. Although new methodology allows us to measure this type of DNA methylation indirectly, there is no reliable statistical method to test for association with a trait. In my program I will develop such a method motivated by a study comparing the brains of suicide victims and controls who died from accidental or natural causes. Since an individual's methylation pattern is known to vary with age I will also examine how to incorporate age and other non-genetic covariates into the association models so that true epigenetic effects can be found.