Table 1.

Comparison of predictions

Methods	Sensitivity (percentage of effective sites correctly predicted)	Specificity (percentage of ineffective sites correctly predicted)	Positive predictive value (%)	Negative predictive value (%)	Prevalence (percentage predicted to be effective)
Duplex + GC	32	75	36	71	28
Accessibility + duplex	48	81	55	78	28
Duplex	84	49	43	87	61

Duplex-end energies had the best sensitivity, but this was offset by a low selectivity. Overall, the best predictions were made using secondary structure accessibility predictions and duplex-end energy differential. This method had the highest selectivity and the best positive predictive value (reliability of positive predictions).