This should be done preferably by directly genotyping this marker. Currently-available imputation methods are powerful and quite accurate for filling in information on missing common SNPs. [35–39] Even then, further conformation by direct genotyping would be very useful. (In fact, to rule out technical artifact, some have argued that an associated SNP should be genotyped using two different genotyping technologies, or that a second SNP in the region that is in [near-] perfect LD with the associated SNP be genotyped.[5]) Great caution is needed when “replicating” an association by finding an association with a (different) nearby marker: if the new marker does not have perfect or almost perfect LD with the previously discovered one, this cannot be considered replication. Moreover, even for markers with seemingly perfect LD in a given sample, the LD may be far less than perfect in a different population and it may break completely in populations of different ancestry. When a panel of markers spanning the whole locus is pursued (e.g. after resequencing and fine mapping), different markers and haplotypes may be found to be associated
