Skip to main content

Table 10 Bootstrap resampling performance analysis of deep-learning based machine learning methods on unbalanced dataset (1:10)

From: Struct2Graph: a graph attention network for structure based predictions of protein–protein interactions

Method

Performance (%) on Unbalanced training set - 1:10

Accuracy

Precision

Recall

Specificity

Decision trees

96.63 ± 0.46

91.66 ± 1.47

73.14 ± 2.93

99.26 ± 0.14

Random forest

97.85 ± 0.25

95.87 ± 0.68

82.12 ± 2.28

99.61 ± 0.06

DeepPPI

98.09 ± 0.77

95.30 ± 3.86

83.29 ± 7.49

99.58 ± 0.37

DeepFE-PPI

98.50 ± 0.46

96.56 ± 0.38

87.86 ± 5.00

99.66 ± 0.05

Struct2Graph

99.26 ± 0.15

97.04 ± 0.70

95.59 ± 0.73

99.67 ± 0.10

Method

MCC

F1-score

ROC-AUC

NPV

Decision trees

80.13 ± 2.06

81.33 ± 1.99

86.20 ± 1.48

97.07 ± 0.45

Random forest

87.60 ± 1.12

88.44 ± 1.11

99.49 ± 0.22

98.03 ± 0.32

DeepPPI

88.01 ± 4.96

88.69 ± 4.82

96.65 ± 1.60

98.35 ± 0.73

DeepFE-PPI

91.27 ± 2.80

91.92 ± 2.72

99.50 ± 0.20

98.69 ± 0.24

Struct2Graph

95.90 ± 0.60

96.31 ± 0.52

99.54 ± 0.22

99.50 ± 0.12

  1. Bold face numbers indicate the best performance