Negative Correlation of KL distances and False-Negative Prediction Error Rates Produced by GeneMark in Short Fragment Identification
| Model (order 3) | B. subtilis | E. coli | M. genitalium | M. jannaschii | M. pneumoniae |
| False-negative error rate | |||||
| AT | 0.186 | 0.223 | 0.171 | 0.118 | 0.165 |
| T | 0.098 | 0.083 | 0.093 | 0.044 | 0.078 |
| HT | 0.054 | 0.045 | 0.067 | 0.017 | 0.052 |
| KL distance (coding vs. noncoding model) | |||||
| AT | 0.0648 | 0.0619 | 0.0837 | 0.0999 | 0.0883 |
| T | 0.0930 | 0.1102 | 0.0893 | 0.1299 | 0.1149 |
| HT | 0.1318 | 0.1687 | 0.1289 | 0.1909 | 0.1587 |
| R = | −0.961 | −0.931 | −0.770 | −0.896 | −0.905 |
[i] For five genomes the error rates are shown along with the KL distance between corresponding coding and noncoding models of the third order. The correlation coefficients (bottom row) for A. fulgidus, H. influenzae, H. pylori, M. thermoautotrophicum, andSynechocystis, averaged with regard to A-, B-, and C-type models, were equal to −0.97, −0.91, −0.96, −0.96, and −0.99, respectively.