Fig. 6

Boxplots of kinship coefficient estimates for the merged populations (CEU, YRI, CHB and JPT). a: HapMap individuals with PO or UN relationships, b: 1000 genomes individuals with UN relationship. The autosomal chromosomes were concatenated for the HapMap data but the randomly selected three chromosomes 3, 18, and 20 were concatenated for the 1000 genomes data to save computation time. The REAP result for the 1000 genomes data is not available because it did not finish within 300 hours of walltime. Individual pairs with valid estimates for all compared methods are used here. Specifically, the following steps were applied: Step 1. A set of clusters of individuals with similar minor allele distributions is discovered by the K-mean clustering. Step 2. For each cluster, the unknown order parameter is estimated (within-cluster estimation) and then the kinship coefficients for all individual pairs in the cluster are estimated. Step 3. For each pair of clusters, the parameter is estimated (between-cluster estimation) and then the kinship coefficients for all the individuals pairs, i.e. one individual from one cluster and the other individual from the other cluster, are estimated. Step 4. Combine the kinship coefficient estimates from Steps 2 and 3