Finally, while we described our imputation approach in terms of two specific scenarios involving the HapMap, it could in fact be generalized to include any number of reference panels of any type (phased/unphased, complete/incomplete) so long as their SNP sets follow a hierarchy such as the ones laid out in Figure 1 and Figure 2. We envision that IMPUTE v2 will be used in a variety of situations. For example, it may soon become standard practice to combine the HapMap Phase II and Phase III datasets to create a compound reference panel like the one in Scenario B, except with all of the reference data phased. Another plausible situation is the version of Scenario B that we described, in which a large set of controls is used to impute genotypes in cases; we discuss some concerns about association testing in this setting in Text S1 and Figure S2. IMPUTE v2 will also be applied in populations beyond the UK controls used in this study, and we expect that its performance will follow trends much like those observed for similar imputation methods [23],[24].
