TABLE 1.
Performance of charged-based and composite predictors
| Data set | n (x + r)b | Results (%) by methoda
|
||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| 11/25
|
Modified 11/25
|
Composite
|
||||||||
| Sens | Spec | Rel | Sens | Spec | Rel | Sens | Spec | Rel | ||
| SI/NSI | 257 (70 + 187) | 90 | 96 | 90 | 76 | 98 | 95 | 84 | 96 | 89 |
| Length variants alone | 44 (25 + 19) | 72 | 100 | 100 | 72 | 100 | 100 | 76 | 95 | 95 |
| ACS | 175 (81 + 94) | 81 | 99 | 99 | 81 | 99 | 99 | 89 | 100 | 100 |
a
For each data set and predictor, the specificity (Spec; proportion of true X4/R5X4/SI sequences correctly predicted by method), sensitivity (Sens; proportion of true R5/NSI sequences predicted), and reliability (Rel; proportion of predicted X4/R5X4/SI sequences actually having that phenotype) are presented. Length variants (V3 sequences not having 35 amino acids) from the combined training set are also analyzed separately.
b
Results represent the total number (n) of sequences (number of CXCR4-using or SI [x] and number of pure CCR5-using or NSI [r] sequences).