Second, we evaluated the effect of varying input genotyped SNP set sizes on overall imputation quality for the intersection-based strategy using the European American and African American control subjects from SAGE. Figure 3 compares the average R2 by MAF for imputed SNPs, resulting from a range of input genotyped SNP sets available from Illumina 1M and its intersections with one or more other arrays. The imputation procedures were iteratively repeated following the removal of 1M SNPs not available on various Illumina arrays (HumanOmni1-Quad, Human660W, 550v1, and HumanHap 300-Duo version 2) or Affymetrix 6.0.
