Table 1.

BUSTED Results on Simulated Data from Lu and Guindon (2014).

Simulation Set	N	ps	pb	Detection Rate, by Method
				NYbs	LGc	BUSTED	BUSTEDFG
Tree 1 α XX	16	0	0	0.020	0.026	0.016	0.020
16 XX	16	0	0	N/A	0.028	0.006	N/A
32 XX	32	0	0	N/A	0.052	0.008	N/A
64 XX	64	0	0	N/A	0.150	0.012	N/A
Tree 1 α XV	16	0.20	0.03	0.102	0.028	0.008	0.259
Tree 2 $\delta \gamma$ XV	10	0.20	0.12	0.368	0.036	0.043	0.058
16 B XU	16	0.20	0.25	0.785	0.122	0.290	0.850
16 B XW	16	0.40	0.25	—	0.498	0.372	0.952
32 B XU	32	0.20	0.10	0.754	0.020	0.540	0.930
32 C XW	32	0.40	0.25	—	0.986	0.660	0.990
64 D XW	64	0.40	0.25	—	1.000	0.656	0.996

Note.—Simulation notation is adopted from the same manuscript; all alignments contained 300 codons, and each simulation scenario generated 500 replicates. N, number of sequences; p_s, proportion of sites in the alignment which were simulated with $\omega >1$ along some of the branches; p_b, proportion of branches in the alignment which were simulated with $\omega >1$ along some of the sites; NYbs, the current version of the Nielsen–Yang branch site models; LGc, the covarion model test of Lu and Guindon (2014); BUSTED, our method testing all branches at once; BUSTED^FG, our method testing the branches designated as foreground during the simulations. For NYbs and BUSTED^FG, the correct foreground branches were selected, whereas for the other two methods, the test is done across all branches. Rates for NYbs and LGc methods are taken from Lu and Guindon (2014), “—” indicates that the test was not performed there, and “N/A” indicates that no foreground branches were specified in the simulation. Note that, for the first four rows where $p_s=p_b=0$, the detection rate corresponds to a false positive rate from a simulation without positive selection; the other rows report true positive rates from a power simulation.