If a study sample consists of individuals with well-defined ancestry similar to one of the reference panels, then it is appropriate to use that one reference panel (the “single best” approach). However, if the study sample consists of individuals with ancestry partially similar to more than one of the reference panels, as may occur by admixture, then the choice of reference panel(s) is not as straightforward. Suggested methods for reference panel selection for this latter situation include the “cosmopolitan” and the “weighted mixture” approaches. The “cosmopolitan” approach combines all available reference data (e.g., HapMap data) into a single reference panel [de Bakker et al., 2006]. The “weighted mixture” approach involves the generation of mixtures of the available reference data. Weights can be determined empirically in order to maximize coverage [Pemberton et al., 2008] or imputation accuracy [Huang et al., 2009]. Alternatively, weights can be specified to match estimates of admixture proportions, which may outperform the cosmopolitan approach because greater weight is given to the HapMap reference panels with more similar ancestry to the study sample [Egyud et al., 2009].
