Table 4.

Exon Level Per Transcript Pair Comparison to Manual Annotations

	Exact pairs	Sn	Sp	Exact pairs (coding)	Sn (coding)	Sp (coding)
chr13	61%	0.40	0.53	80%	0.45	0.75
chr14	64%	0.39	0.60	80%	0.41	0.79
chr20	62%	0.39	0.57	79%	0.43	0.76
chr6	61%	0.38	0.56	79%	0.42	0.75

Only exons that are part of a transcript pair are compared with each other. An annotated exon is considered as found it there is at least one predicted exon with exact matching boundaries in a given transcript pair. The second and fourth columns give the percentage of exact exon matches from the total number of exon pairs formed within the transcript pairs for all exons and for coding exons, respectively. The sensitivity (Sn) is computed as the number of exact exon matches within transcript pairs over the total number of annotated exons within transcript pairs. Similarly, the specificity (Sp) is computed as the number of exact exon matches within transcript pairs over the total number of predicted exons within transcript pairs. These values are given for coding exons only and for coding and noncoding exons together. The percentage of exact matches is higher when restricted to transcript pairs, and it is also higher for coding exons only. The specificity is much higher than in Table 3, and sensitivity is of comparable magnitude, which indicates that the predicted transcripts mostly miss exons rather than overpredict them. The high specificity also indicates that the transcript structures tend to agree with the annotations.