Table 2

Recall of TAIR9 and TAIR10 annotations

TAIR9

TAIR10

Recall

Novel

Transcripts

Recall

Method

Exons

Introns

Exons

Introns

Number

Percentage

Exons

Introns

Splicegrapher

1.00

1.00

1,428

1,282

28

1.4%

0.039

0.045

Splicegrapher + EST

1.00

1.00

11,299

3,557

38

1.9%

0.050

0.056

Cufflinks

No gene models

0.25

0.19

33,252

3,425

0

0.0%

0.035

0.017

With gene models

0.94

0.89

12,690

5,222

4

0.2%

0.017

0.008

TAU

No gene models

0.21

0.29

86,346

5,335

0

0.0%

0.043

0.029

With gene models

0.69

0.79

115,130

3,734

11

1.1%

0.079

0.074

Comparison of the ability of SpliceGrapher, Cufflinks, and TAU to predict TAIR9 and TAIR10 annotations in A. thaliana. The columns of the table provide recall levels of TAIR9 annotations, the number of novel exons and introns that are predicted, i.e., are not in the TAIR9 annotations, the number and percentage of TAIR10 transcripts that are predicted, and the recall level of TAIR10 annotations at the exon and intron level. When TAU and Cufflinks rely on RNA-Seq data alone they tend to produce graphs that are missing many of the features found in the gene models, as reflected in their recall scores between 0.19 and 0.29. When we provide them with TAIR9 annotations their recall scores improve, though TAU's improved statistics result from a vast number of novel exons, many of which may be spurious. When comparing these predictions with novel splice forms in the TAIR10 gene models, SpliceGrapher predicts more novel splice forms correctly than the other packages.

Rogers et al. Genome Biology 2012 13:R4   doi:10.1186/gb-2012-13-1-r4