One of the SNP filters that we recommend is based on HWE test p-value. Interpretation of these p-values is difficult because the choice of significance level depends on sample size [Wakefield 2009]. However, the purpose of the recommended filter is to flag poorly performing assays rather than detecting real deviations in the population, so we examine genotype cluster plots to set a threshold for filtering. In all four studies described here, these plots show that many assays with p-values between 10−3 and 10−4 have good clustering and genotype calling whereas many of those with p-values less than 10−4 are of poor quality. (For example, in the Lung Cancer project, among 48 plots in the range of p=10−6 to 10−4, 12 of 48 plots showed good clustering, whereas in the range of p=10−4 to 10−2, 42 of 48 showed good clustering.) Therefore, we recommend filtering at p=10−4 for these four studies. Other studies may require a different threshold to account for variations in sample size and genotyping technology.
