Skip to main content

Table 1 Comparison of prediction performances of different softwares on three randomly chosen genomic sequences

From: SpliceFinder: ab initio prediction of splice sites using convolutional neural network

(a)

Genomic Sequence I

Donor

Acceptor

Donor & Acceptor

GeneSplicer

N/A

N/A

N/A

SpliceMachine

159

62

221

SpliceRover

19

16

35

SpliceFinder

7

5

12

Genomic Sequence II

Donor

Acceptor

Donor & Acceptor

GeneSplicer

N/A

N/A

N/A

SpliceMachine

72

54

126

SpliceRover

3

6

9

SpliceFinder

1

2

3

Genomic Sequence III

Donor

Acceptor

Donor & Acceptor

GeneSplicer

N/A

N/A

N/A

SpliceMachine

N/A

N/A

N/A

SpliceRover

N/A

N/A

N/A

SpliceFinder

24

35

59

(b)

Genomic Sequence I

Donor

Acceptor

Donor & Acceptor

GeneSplicer

N/A

N/A

N/A

SpliceMachine

159

62

221

SpliceRover

19

3

22

SpliceFinder

5

5

10

Genomic Sequence II

Donor

Acceptor

Donor & Acceptor

GeneSplicer

9

4

13

SpliceMachine

10

4

14

SpliceRover

0

3

3

SpliceFinder

1

2

3

Genomic Sequence III

Donor

Acceptor

Donor & Acceptor

GeneSplicer

N/A

N/A

N/A

SpliceMachine

20

90

110

SpliceRover

444

21

465

SpliceFinder

6

6

12

  1. The numbers of false positives when recall reaches 1 (a) and 0.8 (b) are shown. (Note: N/A implies the software can not reach the recall of 1 or 0.8, no matter how to set the parameters. The best performance is in bold.)