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

BMC Bioinformatics

Table 2 Bootstrap resampling performance analysis of several machine learning methods on balanced dataset (1:1)

Method	Performance (%) on Balanced training set—1:1
Method	Accuracy	Precision	Recall	Specificity
GaussianNB	71.90 ± 3.19	99.01 ± 0.38	46.39 ± 5.95	99.47 ± 0.23
QDA	54.46 ± 0.93	53.28 ± 0.77	99.96 ± 0.06	5.30 ± 1.74
k-NN	92.77 ± 0.22	99.38 ± 0.13	86.62 ± 0.50	99.41 ± 0.11
Decision trees	95.15 ± 0.51	97.13 ± 0.38	93.41 ± 0.88	97.03 ± 0.31
Random forest	98.70 ± 0.11	99.39 ± 0.18	98.10 ± 0.14	99.35 ± 0.18
Adaboost	96.61 ± 0.25	97.83 ± 0.40	95.60 ± 0.21	97.71 ± 0.44
SVC	98.05 ± 0.19	99.35 ± 0.15	96.87 ± 0.28	99.32 ± 0.15
DeepPPI	97.16 ± 0.40	98.04 ± 0.96	96.27 ± 0.84	98.06 ± 0.98
DeepFE-PPI	98.41 ± 0.12	98.81 ± 0.50	98.11 ± 0.45	98.73 ± 0.55
Struct2Graph	98.96 ± 0.19	99.40 ± 0.09	98.57 ± 0.35	99.47 ± 0.09

Method	MCC	F1-score	ROC-AUC	NPV
GaussianNB	53.40 ± 4.44	62.93 ± 5.81	96.04 ± 0.12	63.30 ± 2.72
QDA	16.43 ± 2.53	69.50 ± 0.66	52.63 ± 0.84	99.37 ± 0.83
k-NN	86.36 ± 0.40	92.56 ± 0.30	98.32 ± 0.16	87.31 ± 0.38
Decision trees	95.23 ± 0.55	95.27 ± 0.54	94.21 ± 0.71	93.18 ± 0.81
Random forest	97.41 ± 0.22	98.74 ± 0.11	99.58 ± 0.07	97.97 ± 0.17
Adaboost	93.25 ± 0.50	96.70 ± 0.23	99.00 ± 0.09	95.36 ± 0.29
SVC	96.13 ± 0.37	98.10 ± 0.20	99.53 ± 0.09	96.71 ± 0.28
DeepPPI	94.36 ± 0.81	97.14 ± 0.40	99.05 ± 0.14	96.34 ± 0.78
DeepFE-PPI	96.82 ± 0.25	98.45 ± 0.12	99.52 ± 0.05	97.99 ± 0.20
Struct2Graph	97.91 ± 0.38	98.98 ± 0.19	99.62 ± 0.17	98.50 ± 0.33

Bold face numbers indicate the best performance
Note that the proposed Struct2Graph method outperforms all other methods on the majority of metrics

ISSN: 1471-2105