Quantification and correction of bias in tagging SNPs caused by insufficient sample size and marker density by means of haplotype-dropping.

Tagging single nucleotide polymorphisms (tSNPs) are commonly used to capture genetic diversity cost-effectively. It is important that the efficacy of tSNPs is correctly estimated, otherwise coverage may be inadequate and studies underpowered. Using data simulated under a coalescent model, we show that insufficient sample size can lead to overestimation of tSNP efficacy. Quantifying this we find that even when insufficient marker density is adjusted for, estimates of tSNP efficacy are up to 45% higher than the true values. Even with as many as 100 individuals, estimates of tSNP efficacy may be 9% higher than the true value. We describe a novel method for estimating tSNP efficacy accounting for limited sample size. The method is based on exclusion of haplotypes, incorporating a previous adjustment for insufficient marker density. We show that this method outperforms an existing Bootstrap approach. We compare the efficacy of multimarker and pairwise tSNP selection methods on real data. These confirm our findings with simulated data and suggest that pairwise methods are less sensitive to sample size, but more sensitive to marker density. We conclude that a combination of insufficient sample size and overfitting may cause overestimation of tSNP efficacy and underpowering of studies based on tSNPs. Our novel method corrects much of this bias and is superior to a previous method. However, sample sizes larger than previously suggested may be required for accurate estimation of tSNP efficacy. This has obvious ramifications for tSNP selection both in candidate regions and using HapMap or SNP chips for genomewide studies.