Fig. 1

Power estimates in the presence of a confounder. Plot of estimated power vs. true fold change differences (effect size) using a significance threshold of 0.05 under a gene set simulation with V I F=20.024 and 50 % of probes are confounded. QuSAGE’s power curve is biased to the left and does not maintain a minimum at zero. Q-Gen’s adjustments appropriately controls for type-I error rate and its statistical inference can be trusted when true differences occur