By our measure of information content, low-frequency SNPs have the greatest potential to be informative. In fact, the 10 most informative SNPs have less than 20% heterozygotes in the European American sample. However, these low frequency SNPs also have great potential to provide information that is useless or misleading with regard to the ancestry of an individual. The vast majority of SNPs with a low percentage of heterozygotes provide little or no useful information. As a result, high frequency SNPs, on average, seem to provide better information. This is a likely explanation for why the results in Figure 1 show it requires far fewer SNPs of high MAF to resolve population structure effectively in this sample.
