data, however the degree of across-population replication was higher for the ‘REDUCED’ data, which provides confidence that the method is not simply adding false positive associations. Indeed, this demonstrates that by applying dimension reduction, we do not introduce bias, rather we increase power and replication. For associations detected in more than one population, we find nearly perfect concordance of allelic direction across populations, and find that the absolute allelic effect size, as estimated by the fold-change between the two homozygote classes, remains the same across populations, suggesting little in the way of modification of eQTL effects across populations. The last two results support the idea that while eQTLs may explain different degrees of population variance depending on their frequency in each population [21], the absolute effects on each individual are the same. Given that many GWAS signals are likely to be eQTLs [5], [6], it is likely that many of the GWAS signals discovered in one population may be transferable at the level of the absolute individual risk to other populations. One final observation is that we were able to refine our signals at individual loci to identify putative independently-acting cis-eQTLs.
