Common SNPs (MAF > 5%) were selected, and LDs (r² values) between every SNP pairs within 300 kb were calculated. LD bins were collected so that every pairs within the bin shows r² >0.8, by the greedy method of ldSELECT (Carson et al., AJHG 74, 106-120, 2004) after the program was modified so that it can handle haploid data.

Common SNPs (MAF > 5%) were selected, and LDs (r² values) between every SNP pairs within 300 kb were calculated. LD bins were collected so that SNP pairs within the bin shows r² >0.8, by the greedy method of ldSELECT (Carson et al., AJHG 74, 106-120, 2004) after the program was modified so that it can handle haploid data. tagSNP is the SNP that shows r² >0.8 for all other members of the bin. Best tagSNP is the SNP that shows the highest mean r² for all other members of the bin.

The analyses by Perlegen 281K arrays are described in Kukita et al. (2005). The analyses by Affymetrix 500K are described in Higasa et al. (2009). Affymetrix markers were mapped to NCBI Build 35 (Higasa mapping was on Build 36), and merged with Perlegen data. The original Affymetrix data contained a small fraction (< 1%) of SNPs that were genotyped heterozygote. These typings were judged to be errors because the materials were presumed to be homozygotes. Thus, all heterozygous calls (in Affymetrix data) were forced to be no calls. Many of the heterozygous calls were also low S/N calls and frequently irreproducible when tested. See "LD bin List" for the calculation of LD bins.