The AJHG discusses an interesting phenomenon that I noticed too. While replicating a disease marker association the risk allele suddenly reversed which led me believe that there was an error in allele coding. As the authors of this new paper now show these flip-flop association may indeed a true positive association to a non causal variant.