Table 1.

Percentage of correctly predicted human exons and introns in coding regions grouped by their posterior probability

Posterior probability P	Exon specificity
0 ≤ P ≤ 50%	46/242 ≈ 19.0%
50% < P ≤ 70%	132/356 ≈ 37.1%
70% < P ≤ 80%	84/175 ≈ 48.0%
80% < P ≤ 90%	171/275 ≈ 62.2%
90% < P ≤ 95%	140/195 ≈ 71.8%
95% < P ≤ 99%	338/422 ≈ 80.1%
99% < P ≤ 100%	545/612 ≈ 89.1%
Total	1456/2277 ≈ 63.9%

For example, out of 2277 exons predicted by AUGUSTUS, 422 had a posterior probability between 0.95 and 0.99 of which 338 (80.1%) matched exactly an annotated exon. Here, AUGUSTUS was set to predict only one transcript per gene (no alternatives). As reference annotation the ENCODE test set with 296 genes and 649 transcripts was used, which is a challenging test set: the exon-level specificities of AUGUSTUS, GENEID, GENEZILLA and GENSCAN are 63.9, 61.1, 50.3 and 46.4%, respectively.