Figure 8

Flowchart describing FWER (k = 1) and gFWER (k > 1) algorithms. Genome-wide bootstrap iterations are used to control for multiple testing error. Bootstrap samples of the phenotype vector are used to simulate the random background distribution for association score profiles. For each bootstrap sample, an association method shown in Figure 1 is used to compute one observation of the background distribution. For computational efficiency, gFWER calculations only utilize 500,000 primary and 101 secondary bootstrap samples when estimating significance (Figure 6). Additionally, instead of reporting the median (Figure 6) the maximum is reported to satisfy the hypothesis of the null dominance criterion [23]. When k = 1, this flowchart describes the FWER algorithm. For k > 1, this procedure describes the gFWER approach outlined by Dudoit et al.