Skip to main content

Table 9 CASP7 results, [12, 23] residue separation

From: Ab initio and template-based prediction of multi-class distance maps by two-dimensional recursive neural networks

template ID

[0,10)%

[10,20)%

[20,30)%

[30,40)%

[40,50)%

[50,60)%

[60,70)%

[70,80)%

All

proteins

29

29

16

4

7

4

3

1

93

positive pairs

3676

2336

3006

818

800

804

268

176

11884

negative pairs

117104

95092

84186

27010

26008

25896

7484

6148

388928

SAM_T06

13.6%

11.7%

18.1%

21.5%

16.7%

15.1%

12.7%

12.2%

14.6%

Betapro

18.5%

18.3%

24.4%

34.3%

20.4%

17.6%

25.9%

10.6%

21.4%

ProfCon

26.3%

26.3%

25.2%

29.9%

25.8%

26.0%

27.5%

28.3%

26.3%

Possum

14.8%

18.3%

24.0%

19.6%

18.1%

5.7%

15.3%

18.2%

17.8%

SVMcon

22.1%

22.3%

20.8%

21.5%

23.1%

22.0%

20.0%

40.8%

22.2%

8 AI

25.9%

25.8%

25.7%

40.7%

34.8%

35.6%

31.0%

47.3%

28.5%

8 TE

24.0%

40.2%

64.9%

82.8%

76.9%

91.3%

91.8%

86.5%

52.6%

  1. Results for the CASP7 targets: sequence separations of 12 to 23 residues, inclusive. Comparison between our two predictors (8 AI and 8 TE ) and the predictors ranked highest at CASP7. We report F1 for the contact class, as a function of sequence similarity to the best PSI-BLAST template. Predictions are from the CASP7 web site. ProfCon is in italics because it predicted 73 out of 93 maps, hence its results are not exactly comparable. In bold is the highest F1 of all predictors, excluding ProfCon, and excluding 8 TE , which uses templates and has the highest F1 in all ranges except [0,10)% where it is slightly worse than only 8 AI .