However, the proportion of significant windows was larger for non-replicated (78%) than for replicated (62%) and random genomic SNPs (66%). Figure 4 shows the cumulative distributions of empirical p-values for the three groups of SNPs. The cumulative distribution of p-values correspondent to replicated SNPs does not depart from genome-wide expectations, while non-replicated SNPs clearly map into regions of the genome with extreme differences in LD between Europeans and East Asians. In other words: our observations on LD differences suggest that a proportion of associations would have failed to replicate in East Asians because of population heterogeneity in LD between causal variants and tagSNPs. Yet, the possibilities of heterogeneity across populations in the effect size of causal variants themselves (see [22], [32]) or the presence of European-specific causal rare variants in some associations cannot be discarded as a source of lack of replication.
