Fig. 3

False positive rate plots for both the LRT and Wald tests. Actual false positive rates were calculated at 5 different sample sizes, 5 different mean expression levels, 5 different dispersion values for four tests (LRT under the BNB model, Wald test under the BNB model, LRT under the Poisson-LMM model, LRT under the NB-LMM model). The nominal false positive rate α is set at 0.001 (dashed black lines). The Asymptotic Chi-square distribution is used instead of the empirical parametric test for calculating the critical values. Actual false positive rates of the LRT under the BNB model and the LRT under the Poisson-LMM model are well maintained at the nominal false positive rate. Actual false positive rates of the Wald test under the BNB model are much larger than the nominal false positive rate especially at smaller mean expression levels. Actual false positive rates of the LRT under the NB-LMM model are uniformly lower than the nominal false positive rate