Table 3.
Ab initio gene prediction accuracy results of WebAUGUSTUS with web-trained and expert-trained parameters
| Scenario | Trainer | Gene level |
Exon level |
Nucleotide level |
|||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Sens. | Spec. | #Anno | #Pred | Sens. | Spec. | #Anno | #Pred | Sens. | Spec. | #Anno | #Pred | ||
| optA | web | 46.3 | 37.2 | 5660 | 7099 | 67.6 | 57.8 | 24 846 | 29 082 | 90.3 | 69.8 | 9 387 473 | 12 157 611 |
| Expert | 49.0 | 40.3 | 6897 | 71.5 | 56.8 | 31 241 | 93.2 | 67.0 | 13 053 363 | ||||
| optB | Web | 56.8 | 45.7 | 13 535 | 16 843 | 82.0 | 72.5 | 73 625 | 83 234 | 96.4 | 76.1 | 16 504 394 | 20 906 994 |
| Expert | 58.9 | 46.4 | 16 920 | 83.1 | 71.2 | 85 883 | 96.9 | 75.3 | 21 226 128 | ||||
| optC | Web | 37.2 | 39.3 | 9992 | 9450 | 74.8 | 76.0 | 63 286 | 62 301 | 90.0 | 87.1 | 12 394 167 | 12 791 466 |
| Expert | 32.4 | 36.8 | 8794 | 71.7 | 75.6 | 60 111 | 87.7 | 87.5 | 12 420 039 | ||||
Accuracy was measured by comparing predicted genes to existing annotations. Parameters were optimized using the three different approaches that are available at WebAUGUSTUS: training AUGUSTUS with gene structures that were generated in a fully automated way from ESTs (optA, D. melanogaster) or protein (optB, A. thaliana) sequences, and training AUGUSTUS with externally generated gene structures (optC, C. elegans). For each scenario, we show accuracy results that were obtained using WebAUGUSTUS, and in a row below, the accuracy results obtained with already existing parameter sets that were generated by experts.
Spec., Specificity; Sens., Sensitivity; #Anno, number of annotated features; #Pred, number or predicted features.