Table 2

Comparative evaluation of splice junction prediction tools

75 bp

200 bp

Tool

Sensitivity

Precision

Sensitivity

Precision


CRAC

79.43

99.5

86.02

99.18

GSNAP

84.17

97.03

72.94

97.09

MapSplice

79.89

97.68

84.72

98.82

TopHat

84.96

89.59

54.07

94.69

TopHat2

82.25

92.71

88.65

91.35


We compared the sensitivity and precision of different tools on the human simulated RNA-seq (42M, 75 nt and 48M, 200 nt) against the human genome for splice junction prediction. The sensitivity is the percentage of correctly reported cases over all sequenced cases, while the precision is the percentage of correct cases among all reported cases. Values in bold in the three tables indicate the maximum of a column, and those in italics the second highest values. For all tasks with the current read length, CRAC combines good sensitivity and very good precision. Importantly, CRAC always improves sensitivity with longer reads, and yields the best precision (that is the fewer false positives) over all solutions, even against specialized tools like TopHat.

Philippe et al. Genome Biology 2013 14:R30   doi:10.1186/gb-2013-14-3-r30

Open Data