Fig. 3

Performance of the wLOD method at different SNV densities. Line graphs showing how average power (top), false discovery rate (middle), and ratio of inferred and true ROA length (bottom) across 50 replicate genetic simulations change with increasing ROA length for each SNV subset under (a) scenario 1 and (b) scenario 2. Each comparison was performed at the optimal combination of window size and overlap fraction for that scenario and SNV subset (Additional file 2: Table S1). The grey vertical lines denote 500 kb (dashed) and 1.5 Mb (dotted), frequently applied length thresholds used to categorize ROA arising due to LD (< 500 kb) and inbreeding (> 1.5 Mb) in humans [9]. Note that in scenario1, power to detect ROA > 1 Mb with 18,000 SNVs surpasses that with 50–125,000 SNVs as a consequence of the optimal overlap fraction used: the overlap fraction of 0 used for the 18,000 SNV dataset is much lower than the 0.15–0.22 fractions used for the 50–125,000 SNV datasets. Consequently, greater power to detect ROA > 1 Mb is achieved with 18,000 SNVs than is possible with 50–125,000 SNVs through less stringent placement of ROA boundaries, but at the expense of more frequent overcalling of ROA (inflated false discovery rate)