Table 2. Performance of genotypic algorithms modified to include the “440 rule”.
| Genotypic | Design Sequence Set | Independent Test Sequence Set | ||||
| algorithm | % Sens | % Spec | AUROC | % Sens | % Spec | AUROC |
| 440 rule | 23.1 | 96 | 0.60 | 39.4 | 95.5 | 0.67 |
| 11/25 rule | 65.1 (+9.3) | 99.6 (-3.1) | 0.8 (+0.03; p = 0.30) | 33.3 (+24.3) | 98.5 (−4.5) | 0.66 (+0.10; p = 0.11) |
| G2P FPR 1% | 55.8 (+9.3) | 99.6 (−4.1) | 0.78 (+0.03; p = 0.33) | 33.3 (+24.3) | 100 (−4.5) | 0.58 (+0.18; p = 0.01) |
| G2P FPR 2.5% | 74.4 (+7) | 98.2 (−4) | 0.86 (+0.02; p = 0.38) | 51.5 (+15.2) | 100 (−4.5) | 0.63 (+0.18; p = 0.01) |
| G2P FPR 5% | 79.1 (+6.9) | 94.2 (−3.6) | 0.87 (+0.01; p = 0.37) | 60.6 (+15.2) | 100 (−4.5) | 0.65 (+0.21; p = 0.003) |
| G2P FPR 5.75% | 81.4 (+7) | 92.1 (−3.6) | 0.87 (+0.02; p = 0.37) | 60.6 (+15.2) | 100 (−4.5) | 0.65 (+0.21; p = 0.003) |
| G2P FPR 10% | 81.4 (+7) | 86.2 (−3.6) | 0.84 (+0.02; p = 0.38) | 69.7 (+9.1) | 91 (−4.4) | 0.80 (+0.02; p = 0.37) |
| G2P FPR 15% | 81.4 (+7) | 82.1 (−3.1) | 0.82 (+0.02; p = 0.37) | 69.7 (+9.1) | 85.1 (−4.5) | 0.77 (+0.02; p = 0.43) |
| G2P FPR 20% | 83.7 (+7) | 70.5 (−2.2) | 0.77 (+0.02; p = 0.35) | 72.7 (+9.1) | 77.6 (−4.5) | 0.75 (+0.02; p = 0.38) |
| WebPSSMX4R5 | 74.4 (+7) | 98.2 (−3.6) | 0.86 (+0.02; p = 0.74) | 48.5 (+15.1) | 98.5 (−4.5) | 0.74 (+0.05; p = 0.25) |
| WebPSSMSI/NSI | 76.7 (+7) | 94.2 (−3.1) | 0.85 (+0.02; p = 0.71) | 48.5 (+15.1) | 97 (−4.5) | 0.73 (+0.05; p = 0.25) |
The “440 rule” involves rescreening sequences predicted to be R5 by the indicated genotypic algorithm for the presence of Asp or Glu at position 440, the presence of which results in a “CXCR4-using” prediction. % Sens, sensitivity was calculated by dividing the number of correctly predicted CXCR4-using sequences by the total number of phenotypically characterised CXCR4-using sequences. % Spec, specificity was calculated by dividing the number of correctly predicted R5 sequences by the number of phenotypically characterised R5 sequences. Values in parentheses represent percentage differences between modified and unmodified algorithms. Differences in area under the receiver operating characteristic curve (AUROC) considered significant (p≤0.05) are highlighted in bold and italicized. 43 CXCR4-using (23 R5X4 and 20 X4) and 223 R5 B-HIV design sequences and 32 CXCR4-using (23 R5X4 and 9 X4) and 70 R5 B-HIV test sequences were analysed. FPR, false positive rate.