Skip to main content

Table 4 The performance (precision, recall and F-score) of six GPI extraction methods when applied to five GPI corpora using gold-standard named entities.

From: A realistic assessment of methods for extracting gene/protein interactions from free text

 

A

B

H

I

L

Precision:

     

AkanePPI(A)

(57.0)

29.2

61.5

60.2

69.6

AkanePPI(B)

29.1

(56.8)

52.0

66.2

76.7

RelEx

40

39

76

74

82

Baseline(K)

22.8

24

54

44.8

(53.9)

Baseline(C)

17

13

38

41

50

OpenDMAP

61

62.3

77.3

87.5

100

Recall:

     

AkanePPI(A)

(74.0)

31.8

44.2

32.5

23.8

AkanePPI(B)

52.9

(85.4)

55.8

51.3

40.2

RelEx

50

45

64

61

72

Baseline(K)

51.5

52.2

66.9

56.4

(72)

Baseline(C)

95

99

100

100

100

OpenDMAP

9.1

5.9

10.4

2.1

2.4

F-score:

     

AkanePPI(A)

(64.4)

30.5

51.4

42.2

35.4

AkanePPI(B)

37.5

(68.2)

53.8

57.8

52.8

RelEx

44

41

69

67

77

Baseline(K)

31.6

32.9

59.7

49.9

(61.6)

Baseline(C)

29

23

55

58

66

OpenDMAP

15.9

10.8

18.4

4.1

4.8

  1. The figures for RelEx and Baseline(C) are taken from Pyysalo et al. (2008). (Note that we use a simplified version of BioInfer compared to the one used in that paper, so the figures for this corpus are not completely comparable.) Figures are given in brackets where a corpus was used to develop a given method. Corpus abbreviations are as follows: A = AIMed; B = BioInfer; H = HPRD50; I = IEPA; L = LLL.