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

BMC Bioinformatics

Table 8 Bootstrap resampling performance analysis of deep-learning based machine learning methods on unbalanced dataset (1:3)

Method	Performance (%) on Unbalanced training set—1:3
Method	Accuracy	Precision	Recall	Specificity
Decision trees	95.72 ± 0.46	95.55 ± 0.74	87.98 ± 1.63	98.52 ± 0.26
Random forest	98.80 ± 0.14	98.29 ± 0.33	97.15 ± 0.34	99.39 ± 0.11
DeepPPI	97.78 ± 0.36	98.43 ± 0.42	92.60 ± 1.38	99.51 ± 0.13
DeepFE-PPI	98.86 ± 0.11	98.42 ± 0.54	97.30 ± 0.31	99.43 ± 0.20
Struct2Graph	99.01 ± 0.16	98.83 ± 0.37	97.42 ± 0.51	99.59 ± 0.13

Method	MCC	F1-score	ROC-AUC	NPV
Decision trees	88.88 ± 1.14	91.60 ± 0.88	93.26 ± 0.81	95.78 ± 0.61
Random forest	96.90 ± 0.39	97.72 ± 0.29	99.72 ± 0.05	98.98 ± 0.10
DeepPPI	94.04 ± 0.98	95.42 ± 0.77	99.06 ± 0.47	97.58 ± 0.44
DeepFE-PPI	97.08 ± 0.28	97.85 ± 0.20	99.34 ± 0.11	99.03 ± 0.05
Struct2Graph	97.46 ± 0.42	98.12 ± 0.32	99.75 ± 0.20	99.08 ± 0.18

ISSN: 1471-2105