Skip to main content

Table 3 PRANK vs. simpler rescoring methods

From: Improving protein-ligand binding site prediction accuracy by classification of inner pocket points using local features

Dataset

Top-n [%]

All [%]

PRANK [%]

Δ PRANK

PLB [%]

Δ PLB

VOL [%]

Δ VOL

Fpocket predictions

CHEN11

47.9

71

58.8**

+10.6

49.8

+1.9

34.5

-13.4

ASTEX

58

81.1

63.6

+5.6

56.6

-1.4

32.2

-25.9

DT198

37.5

80.2

56.2

+18.8

43.2

+5.7

19.3

-18.2

MP210

56.6

78.8

67.7

+11.1

54.5

-2.1

30.6

-26

B48

74.1

92.6

81.5

+7.4

72.2

-1.9

42.6

-31.5

U48

53.7

88.9

77.8

+24.1

66.7

+13

31.5

-22.2

ConCavity predictions

CHEN11

47.9

52.3

50.7**

+2.8

50.4

+2.5

50.2

+2.3

ASTEX

55.2

65.7

62.9

+7.7

62.9

+7.7

63.6

+8.4

DT198

45.8

65.6

61.5

+15.6

56.8

+10.9

59.4

+13.5

MP210

57.4

68.2

66.1

+8.7

64.9

+7.3

64.6

+6.9

B48

66.7

81.5

77.8

+11.1

79.6

+13

75.9

+9.3

U48

64.8

77.8

74.1

+9.3

75.9

+11.1

70.4

+5.6

  1. PLB - rescoring by the Propensity for Ligand Binding index based on amino acid composition of pockets [29].
  2. VOL - rescoring by approximate volume.
  3. **cross-validation results.
  4. The number presented for rescoring methods (columns: PRANK,PLB,VOL) is the success rate considering Top-n predicted pockets measured by DCA criterion with 4 Å threshold.