Figure 2. Power loss from the inclusion of non-causal SNPs with complex LD structure using a window approach to include inter-genic SNPs.

Figure 2 illustrates the power loss observed for all five tests in the realistic LD structure simulation, as more and more non-causal SNPs are added to the test. In particular, a notable decline in power is noted for all tests as the number of intra-genic and inter-genic non-causal SNPs increases. The notable exception is when a small increase in power is observed for most tests when a small region (window) outside of the gene is included since two causal SNPs are located nearby to the gene. Power loss from the inclusion of non-causal SNPs can be substantial, in particular the inclusion of intra-genic non-causal SNPs substantially decreases power. Thus, using increasingly large windows only decreases power when no causal SNPs are located far from the gene.