These considerations apply to imputation datasets in general, but it is particularly interesting to examine them in the context of multiple reference panels genotyped on different sets of SNPs. BEAGLE's joint approach to such datasets is flexible, but we have seen that it can lose accuracy when certain kinds of new data are added. Conversely, IMPUTE v2's multi-panel modeling strategy responds intuitively to new sources of information like additional individuals or SNPs. This property makes it easy to predict how IMPUTE v2 will perform in larger and more complex datasets than the ones used here, whereas the same cannot necessarily be said for BEAGLE.
