Table 2.

Sites Identified by BUSTED, FEEDS, MEDS, and EDEPS in HIV-1 RT.

Site	BUSTED Evidence Ratio	MEDS P Value	FEEDS P Value	EDEPS Bayes Factor	Resistance
41	—a	0.00259	—	—	NRTIb
62	—	—	—	313	NRTI accessory
64	11.9612	0.00244	0.0067	—	NRTI accessory
65	7.36119	—	—	—	NRTI
69	9.70622	—	—	—	NRTI accessory
75	11.7751	—	—	—	NRTI accessory
77	—	—	—	211	NRTI
98	—	0.00488	—	—	—c
100	—	$<0.0001$	—	$>{10}^5$	NNRTId
102	—	—	0.0025	—
103	74.6745	<0.0001	<0.0001	$>{10}^5$	NNRTI
104	—	0.00244	—	—
115	—	—	—	3,110	NRTI
116	—	0.00319	—	—	NRTI accessory
151	24.1913	<0.0001	—	$>{10}^5$	NRTI
162	—	—	—	1,772
165	—	<0.0001	—	2,245
174	—	—	—	105
181	34.5139	<0.0001	—	$>{10}^5$	NNRTI
184	40.6307	<0.0001	—	$>{10}^5$	NRTI
188	29.4723	<0.0001	0.0002	$>{10}^5$	NNRTI
190	15.8163	<0.0001	—	$>{10}^5$	NNRTI
200	10.8219	—	<0.0001	—
215	14.8404	0.00035	—	2,727	NRTI
219	7.32553	—	—	—	NRTI
228	13.38	0.00029	—	1,401	NRTI accessory
230	—	0.00297	—	$>{10}^5$	NNRTI
245	17.9176	—	0.0006	—
286	—	0.00085	—	—

Note.—All methods employ the same foreground partition. FEEDS uses a site-wise fixed effects likelihood ratio test with distinct ω values estimated along FG and BG branches for each site, but admitting no stochastic branch-to-branch variation. MEDS and EDEPS are specifically designed with HIV-1 drug resistance in mind, powered to detect elevated substitution rates toward specific target residues at individual sites, using either a fixed effects codon approach (MEDS) or a site-wise random effects amino acid model (EDEPS). It is remarkable that introducing stochastic variation among branches (i.e., BUSTED) achieves similar power to explicitly directional methods on a genuinely directional system, even when directionality is not explicitly modeled.

^aNot significant.

^bNucleoside reverse-transcriptase inhibitor.

^cMEDS identifies 98S, but only 98G is a resistance mutation.

^dNonnucleoside reverse-transcriptase inhibitor.