. 2015 Feb 25;5:8543. doi: 10.1038/srep08543

Table 1. Sensitivity, specificity and AUROC of genotypic algorithms for predicting coreceptor usage of training set V3 sequences.

		PhenoSeq	g2p FPR 5.75%	g2p FPR 10%	WebPSSM_X4R5	WebPSSM_SI/NSI^*
PhenoSeq-B	Sens/Spec	80.7/82.9	68.8/93.7	75.3/86.6	53.8/95.8	61.3/94.7
Training Set	AUROC	0.82	0.81 (p = 0.40)	0.81 (p = 0.40)	0.75 (p = 0.05)	0.78 (p = 0.17)
93 CXCR4-using, 269 R5
PhenoSeq-C	Sens/Spec	88.8/87.9	75/96.5	81.3/93.7	-	82.5/89
Training Set	AUROC	0.88	0.86 (p = 0.24)	0.88 (p = 0.40)	-	0.86 (p = 0.24)
80 CXCR4-using, 429 R5
PhenoSeq-D	Sens/Spec	87.8/87.5	81.4/72.7	88.4/61.4	81.4/65.9	69.8/63.6
Training Set	AUROC	0.88	0.77 (*p = 0.02*)	0.72 (*p < 0.01*)	0.72 (*p = 0.03*)	0.77 (*p = 0.02*)
57 CXCR4-using, 80 R5
PhenoSeq-AE	Sens/Spec	88/92.2	82/76.6	88/56.3	78/78.9	80/81.3
Training Set	AUROC	0.90	0.79 (*p = 0.02*)	0.72 (*p < 0.01*)	0.79 (*p < 0.01*)	0.81 (*p = 0.03*)
50 CXCR4-using, 128 R5
PhenoSeq-A/AG	Sens/Spec	59.7/87.1	29/96	40.3/91.5	27.4/92	29/97
Training Set	AUROC	0.73	0.63 (*p = 0.03*)	0.659 (p = 0.10)	0.60 (*p < 0.01*)	0.63 (*p = 0.04*)
59 CXCR4-using, 172 R5

Sens, % sensitivity for correctly predicting CXCR4-usage (relative to phenotypic tropism assay results) was calculated by dividing the number of correctly predicted CXCR4-using sequences by the total number CXCR4-using sequences and multiplying by 100. Spec, % specificity for correctly predicting R5 strains (relative to phenotypic tropism assay results) was calculated by dividing the number of correctly predicted R5 sequences by the total number of R5 sequences and multiplying by 100. P-values (two-tailed) for comparing area under the receiver operator characteristic curves (AUROC) was performed according to Hanley et al³³. P-values < 0.05 were considered significant and are highlighted in bold italicized text. FPR, false positive rate.

*The subtype C specific WebPSSM_SI/NSI algorithm was used for subtype C HIV-1 predictions.