Detecting rare variant associations by identity-by-descent mapping in case-control studies

Abstract

Identity-by-descent (IBD) mapping tests whether cases share more segments of IBD around a putative causal variant than do controls. These segments of IBD can be accurately detected from genome-wide SNP data. We investigate the power of IBD mapping relative to that of SNP association testing for genome-wide case-control SNP data. Our focus is particularly on rare variants, as these tend to be more recent and hence more likely to have recent shared ancestry. We simulate data from both large and small populations and find that the relative performance of IBD mapping and SNP association testing depends on population demographic history and the strength of selection against causal variants. We also present an IBD mapping analysis of a type 1 diabetes data set. In those data we find that we can detect association only with the HLA region using IBD mapping. Overall, our results suggest that IBD mapping may have higher power than association analysis of SNP data when multiple rare causal variants are clustered within a gene. However, for outbred populations, very large sample sizes may be required for genome-wide significance unless the causal variants have strong effects.

Figure 1

Simulation scheme. Each simulated region is made up of 100 simulated segments of length 1 kb with gaps of length 1 kb between them. The central five segments can contain causal SNPs. Causal SNPs are those that the simulation program designates as protein-changing mutations. These SNPs have been subject to negative selection at a specified rate. Only the causal SNPs and one SNP per segment with highest minor allele frequency (MAF) are retained. The causal SNPs are used to determine disease status, while the high MAF SNPs are tested in the association analysis. IBD status is determined through further simulation, as described in the main text.

Figure 2

Distribution of lengths of detected IBD segments in the WTCCC type 1 diabetes data. IBD segments were detected using BEAGLE fastIBD. Lengths greater than 8 cM are not shown.

Figure 3

Permutation P-values for the IBD test in the WTCCC type 1 diabetes data. P-values were calculated at every tenth marker along the autosomes. The smallest possible P-value from the 5,000,000 permutations (2 × 10⁻⁷) is shown by the black horizontal line. The genome-wide significance level determined by 1000 permutations (6 × 10⁻⁶) is shown by the blue horizontal line.