Table 2. Influences of average relative entropy on annotation for all human families.
| Average relative entropy | Overextension change (bp) | True positive change (bp) |
|---|---|---|
| 0.40 | 1 187 973 | (472 522) |
| 0.42 | 686 251 | (255 520) |
| 0.44 | 242 979 | (43 430) |
| 0.46 | (242 979) | 43 430 |
| 0.48 | (717 069) | 119 880 |
| 0.50 | (1 160 532) | 200 545 |
| 0.52 | (1 586 452) | 104 369 |
| 0.54 | (2 001 938) | (2416) |
| 0.56 | (2 440 201) | (121 426) |
| 0.58 | (2 812 553) | (287 361) |
| 0.60 | (3 236 722) | (586 724) |
| 0.62 | (3 624 644) | (873 070) |
| 0.64 | (3 976 080) | (1 264 674) |
| 0.66 | (4 326 797) | (1 629 825) |
| 0.68 | (4 670 024) | (2 011 169) |
| 0.70 | (4 974 778) | (2 462 143) |
Using the GARLIC benchmark with inserted TE fragments, we tested a variety of target average relative entropy values, assessing the impact on coverage and overextension across all human models. Values in parentheses are negative, indicating a reduction in overextension or coverage from the previous default of 0.45 bits per position. We chose to update the default in HMMER to a higher value (0.62) to reduce overextension while only sacrificing a modest amount of true positive matches.