Description
Consideration of haplotypes of candidate genes is the basis for an effective association study of complex diseases. A problem, however, arises because composite genotypes are the observed data. Haplotypes are not directly observed; they are estimated from genotype data. Generally, haplotypes are maximum likelihood estimates (MLE) established using, for example, an expectation-maximization (EM) algorithm. If such estimated haplotypes are considered and analyzed as if they were directly observed data, without considering the phase uncertainly, inflation of the type I errors may occur and tests become anti-conservative. One solution is to use a likelihood approach to consider all possible haplotypes with their corresponding probabilities. Here we discuss an alternative approach using Monte Carlo testing, and introduce hapMC, a JAVA program, that performs such analyses. The key to the Monte Carlo procedure is to appropriately match the observed statistic and the simulated null statistics that form the null, such that a valid test of the correct size is maintained. This is achieved as follows: MLE haplotypes are established for the observed data (cases and controls separately), and the statistic of interest calculated, ignoring the phase uncertainty. Haplotype frequencies are estimated from the observed data (cases and controls together) and these frequencies used to assign haplotypes to individuals (independent of case/control status). This generates a null configuration of haplotype data. The known phase is then ignored, considered as genotypes and MLE haplotypes are estimated for the null configuration. Again, the MLE haplotypes are estimated in cases and controls separately. The statistic of interest is calculated on the null data. Repeating, a null distribution is produced for testing the observed statistic against. Our program, hapMC, performs the necessary Monte Carlo procedure and EM estimates of haplotypes to provide valid haplotype tests for various standard association statistics, appropriately accounting for phase uncertainty.
|