The effects of errors in the specification of genotype of a single nucleotide polymorphism (SNP) on the chi-squared test of association between disease and SNP type

The purpose of this paper is to study the effects of SNP genotyping errors on the 2 x 3 chi2 test of independence. First, we would like to ask what errors are most costly, in terms of increased sample size to maintain constant asymptotic power and significance level (SSN), when performing case-control studies of genetic association. The most costly error is recording the more common homozygote as the less common homozygote. In general, recording a more common genotype as a less common genotype is more costly than the reverse error with cost increasing as the difference between genotype frequencies increases.; Then we ask what classification rules minimize the SSN when the underlying datum of genotyping follows a three-component normal mixture distribution. That is, for genotype AA, the datum is N( dL,1); for genotype AB, N(0,1); and for genotype BB, N(dR,1), with dL < 0 < dR. The rule to classify an observation as the genotype with closer mean (called the halfway rule) has SSN within 6% of the optimal SSN.; In practice, a "no-call" procedure is sometimes used in which borderline observations are not classified. This procedure has the simultaneous effect of reducing the genotype error rate and the expected number of genotypes observed. Both quantities affect the expected power of the chi-squared statistic. The benefits of reduced genotype error rate are almost exactly balanced by the losses due to reduced genotype observations. For an underlying univariate normal mixture of genotype classification to be analyzed with a 2 x 3 chi-squared test, there is little, if any, increase in power using a no-call procedure.