Skip to main content

Table 3 Three-level LINDAHL fold recognition results of FoldHSphere and FoldHSpherePro in comparison with the state-of-the-art

From: FoldHSphere: deep hyperspherical embeddings for protein fold recognition

Method

Family

Superfamily

Fold

Top 1

Top 5

Top 1

Top 5

Top 1

Top 5

PSI-BLAST [3]

71.2

72.3

27.4

27.9

4.0

4.7

HHpred [14]

82.9

87.1

58.0

70.0

25.2

39.4

RAPTOR [14]

86.6

89.3

56.3

69.0

38.2

58.7

BoostThreader [14]

86.5

90.5

66.1

76.4

42.6

57.4

SPARKS-X [15]

84.1

90.3

59.0

76.3

45.2

67.0

FOLDpro [34]

85.0

89.9

55.0

70.0

26.5

48.3

RF-Fold [34]

84.5

91.5

63.4

79.3

40.8

58.3

DN-Fold [34]

84.5

91.2

61.5

76.5

33.6

60.7

RFDN-Fold [34]

84.7

91.5

65.7

78.8

37.7

61.7

MRFalign [43]

85.2

90.8

72.4

80.9

38.6

56.7

CEthreader [48]

76.6

87.2

69.4

81.8

52.3

70.4

DeepFR (s2) [43]

65.4

83.4

51.4

67.1

56.1

70.1

DeepFRpro (s2) [43]

83.1

92.3

69.6

82.5

66.0

78.8

VGGfold [47]

67.9

84.3

53.2

68.4

58.3

73.5

CNN-BGRU [48]

71.0

87.7

60.1

77.2

58.3

78.8

CNN-BGRU-RF+ [48]

85.4

93.5

73.3

87.8

76.3

85.7

FoldHSphere

76.4

89.2

72.8

86.4

75.1

84.1

FoldHSpherePro

85.2

93.0

79.0

89.2

81.3

90.3

  1. The accuracy (%) results are provided at the family, superfamily and fold levels, considering both the top 1 and top 5 ranked templates. Boldface indicates best performance