Table 3.

Cluster-based prediction of phenotypic resistance relative to controls.

With all data		RTV	NFV	ATV	APV	IDV	LPV	SQV	TPV	DRV
CTL1 (Random)	% correct	35	36	29	21	22	26	29	31	29
CTL1 (Random)	RMSE	1.34	1.13	1.21	1.2	1.05	1.01	1.26	0.98	0.76
CTL2 (Average)	% correct	46	43	62	60	62	56	57	47	34
CTL2 (Average)	RMSE	0.60	0.54	0.36	0.34	0.26	0.28	0.41	0.67	0.67
Cluster-based	% correct	81	75	74	70	63	67	65	50	67
Cluster-based	RMSE	0.35	0.34	0.38	0.33	0.29	0.25	0.50	0.71	0.29
Without nonresistant clusters		RTV	NFV	ATV	APV	IDV	LPV	SQV	TPV	DRV
CTL1 (Random)	% correct	78	66	45	27	22	29	34	18	29
CTL1 (Random)	RMSE	0.54	0.68	0.84	0.97	0.86	0.83	1.06	1.00	0.74
CTL2 (Average)	% correct	28	32	49	51	52	43	46	28	11
CTL2 (Average)	RMSE	0.74	0.64	0.45	0.41	0.32	0.34	0.51	0.84	0.84
Cluster-based	% correct	89	82	73	62	55	58	60	34	56
Cluster-based	RMSE	0.26	0.23	0.38	0.40	0.36	0.31	0.62	0.89	0.36

Percent of viruses whose resistance score toward each drug was correctly classified ("% correct"), as well as the RMS error (in scaled resistance units) over all sequences of the phenotypic difference between predicted and actual phenotype ("RMSE") using the two controls described in the text ("CTL1 (Random)" and "CTL2 (Average)" and the cluster-based prediction. The top panel presents results using all 398 sequences, and the bottom panel shows results after removing the two clusters showing little or no phenotypic resistance to any drug.