One difficulty was how to simulate linkage disequilibrium (LD). We chose to try different approaches to this, including choosing which haplotypes held the disease locus, based on structure and based on frequency. As a model, we used an area of the genome in which there was some marker × marker LD and used the two-SNP haplotype probabilities from that region to simulate the data, modifying the frequencies of the disease allele-carrying haplotypes so that association would be detectable. We tried to calibrate the parameters so that association with SNPs near the disease SNP (which was not one of the ones for which there was typing data) would show statistically significant evidence of association in at least 50% of the datasets. In the end, this calibration was difficult and we ran up against the deadline. Thus, we are less sure about the ultimate product for this aspect.
