Genotyping error rates can be estimated from duplicate discordance rates. Each of the three genotypes may be miscalled as either of the other two genotypes, resulting in six potentially different error rates. For a given true genotype, we consider two error rates, α and β. The probability that duplicate genotyping instances of the same subject give a discordant genotype is 2[(1− α − β)(α + β) + αβ]. When α and β are very small, this is approximately 2(α + β) or twice the total error rate. In high-density genotyping, the number of SNPs per sample is so high that duplicating a single sample would give a good estimate of overall error rate, assuming that the rate was similar for every sample. However, DNA sample quality may vary considerably so that error rates can vary among samples. Therefore, we recommend using at least five study samples for estimating error rates.
