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

BMC Bioinformatics

Table 9 Bootstrap resampling performance analysis of deep-learning based machine learning methods on unbalanced dataset (1:5)

Method	Performance (%) on Unbalanced training set—1:5
Method	Accuracy	Precision	Recall	Specificity
Decision trees	95.56 ± 0.45	94.14 ± 0.79	80.91 ± 2.12	98.87 ± 0.16
Random forest	98.38 ± 0.19	98.25 ± 0.65	92.82 ± 1.07	99.63 ± 0.13
DeepPPI	97.96 ± 0.46	97.13 ± 2.71	90.58 ± 3.21	99.44 ± 0.57
DeepFE-PPI	98.90 ± 0.31	98.20 ± 0.29	95.64 ± 1.75	99.61 ± 0.07
Struct2Graph	99.16 ± 0.17	98.29 ± 0.64	97.08 ± 1.01	99.69 ± 0.13

Method	MCC	F1-score	ROC-AUC	NPV
Decision trees	84.72 ± 1.33	87.01 ± 1.18	89.89 ± 1.04	95.82 ± 0.55
Random forest	94.53 ± 0.70	95.45 ± 0.61	99.66 ± 0.15	99.03 ± 0.11
DeepPPI	92.58 ± 1.66	93.66 ± 1.46	98.96 ± 0.13	98.15 ± 0.61
DeepFE-PPI	96.24 ± 1.11	96.89 ± 0.94	99.49 ± 0.11	99.05 ± 0.16
Struct2Graph	97.03 ± 0.51	97.53 ± 0.41	99.71 ± 0.26	99.40 ± 0.23

ISSN: 1471-2105