Is it rare or common?

Abstract

Many genome-wide association studies (GWAS) have signals with unknown etiology. This paper addresses the question-is such an association signal caused by rare or common variants that lead to increased disease risk? For a genomic region implicated by a GWAS, we use single nucleotide polymorphism (SNP) data in a case-control setting to predict how many common or rare variants there are, using a Bayesian analysis. Our objective is to compute posterior probabilities for configurations of rare and/or common variants. We use an extension of coalescent trees--the ancestral recombination graphs--to model the genealogical history of the samples based on marker data. As we expect SNPs to be in linkage disequilibrium with common disease variants, we can expect the trees to reflect the type of variants. To demonstrate the application, we apply our method to candidate gene sequencing data from a German case-control study on nonsyndromic cleft lip with or without cleft palate.

Figure 1. Two scenarios where a complex disease is caused by a common or two rare variants

Figure 2. Flowchart for the algorithm

Figure 3. A simulated dendrogram for the dataset with cases and controls shown corresponding to the leaves, as red and black dots respectively

Note that some subjects have identical genotypes and therefore are grouped together.

Figure 4. Smoothed posterior for real dataset

The X and Y axes denote counts of rare and common variants, and the Z axis shows the (un-normalized) posterior density. The density is smoothed with a Gaussian kernel.

Figure 5. Posterior for simulated dataset with rare variants

Figure 6. Posterior for simulated dataset with common variants

Figure 7. Posterior for simulated dataset with no variants

Figure 8. Some possible DSL configurations