Results on the Test Sets
| Test set | Nucleotide level | Exon level | ||||||
| SN | SP | AC | SN | SP | (SN+SP)/2 | ME | WE | |
| The ROSETTA set | ||||||||
| ROSETTA | 0.935 | 0.978 | 0.949 | 0.833 | 0.829 | 0.831 | 0.048 | 0.047 |
| SGP-1 | 0.940 | 0.960 | 0.940 | 0.700 | 0.760 | 0.730 | 0.120 | 0.040 |
| SLAM | 0.951 | 0.981 | 0.960 | 0.783 | 0.755 | 0.769 | 0.038 | 0.057 |
| TWINSCAN.p | 0.960 | 0.941 | 0.940 | 0.855 | 0.824 | 0.840 | 0.045 | 0.081 |
| TWINSCAN | 0.984 | 0.889 | 0.923 | 0.839 | 0.767 | 0.803 | 0.034 | 0.118 |
| GENSCAN | 0.975 | 0.908 | 0.929 | 0.817 | 0.770 | 0.793 | 0.057 | 0.107 |
| HoxA | ||||||||
| SLAM | 0.852 | 0.896 | 0.864 | 0.727 | 0.533 | 0.630 | 0.000 | 0.333 |
| TWINSCAN.p | 0.976 | 0.829 | 0.896 | 0.773 | 0.531 | 0.652 | 0.000 | 0.312 |
| TWINSCAN | 0.949 | 0.511 | 0.704 | 0.591 | 0.173 | 0.382 | 0.000 | 0.707 |
| SGP-2 | 0.640 | 0.637 | 0.619 | 0.409 | 0.173 | 0.291 | 0.091 | 0.596 |
| GENSCAN | 0.932 | 0.687 | 0.796 | 0.545 | 0.235 | 0.390 | 0.000 | 0.569 |
| Elastin | ||||||||
| SLAM | 0.876 | 0.981 | 0.926 | 0.802 | 0.859 | 0.831 | 0.121 | 0.059 |
| TWINSCAN.p | 0.942 | 0.950 | 0.945 | 0.879 | 0.889 | 0.884 | 0.066 | 0.056 |
| TWINSCAN | 0.933 | 0.877 | 0.903 | 0.835 | 0.826 | 0.831 | 0.110 | 0.120 |
| SGP-2 | 0.755 | 0.998 | 0.873 | 0.593 | 0.900 | 0.291 | 0.352 | 0.017 |
| GENSCAN | 0.947 | 0.766 | 0.852 | 0.835 | 0.731 | 0.783 | 0.121 | 0.231 |
-
The measures of sensitivity SN = TP/TP + FN and specificity SP = TP/TP + FP (where TP = true positives, TN = true negatives, FP = false positives and FN = false negatives) are shown at both the nucleotide and exon level. ME is entirely missed exons, WE is wrong exons, and the approximate correlation AC = 1/2 (TP/TP + FN + TP/TP + FP + TN/TN + FP + TN/TN + FN) − 1 summarizes the overall nucleotide sensitivity and specificity by one number. Within each of the three data sets the methods are divided into three classes: those operating on a syntenic DNA pair, those operating on a human sequence using as evidence matches against a database of mouse sequences, and a single-organism gene finder (GENSCAN).











