When we generate the trios, we use a roulette wheel algorithm to assign SNPs to the children. This randomness is reflected on the graphs as the dots spread in some irregular patterns. Reading the graphs from left to right, we can see that with low prevalence (K = 0.01) the dots appear in clear clusters. Each cluster corresponds to a specific f 11 value. Taking the top left graph (dominant with K = 0.01) as an example, f 11 changes in the order of 0.9K, 0.7K, 0.5K, 0.3K, 0.1K and 0.0. Within each cluster, f 12 & f 22 increase in the order of 1.1K, 0.5, 0.9 and 1.0. This clustering holds true in the other two disease models (recessive and co-dominant) when K is small (K = 0.01) except some f 11 valued clusters are missing because combinations with invalid p were excluded. In summary, the above observation indicates that f 11 has a higher impact to the power than f 12 & f 22 do in a rare disease model. As the prevalence increases (K = 0.1 and
