A multitask transfer learning framework for the prediction of virus-human protein–protein interactions

BMC Bioinformatics

Table 4 Comparison with DeepViral—a method that can utilize knowledge from the disease phenotype, virus taxonomy, the human PPI network, and the protein GO annotation

Dataset	Model	AUC	AP	Precision	Recall	F1
2697049	DeepViral	0.7288	0.0015	0.07	0.07	0.07
2697049	MTT	0.7566	0.0021	0.97	0.97	0.97
333761	DeepViral	0.8009	0.0147	1.72	1.72	1.72
333761	MTT	0.8160	0.0262	6.35	6.35	6.35
2043570	DeepViral	0.7708	0.0116	0.52	0.52	0.52
2043570	MTT	0.6956	0.0096	1.89	1.91	1.90
644788	DeepViral	0.9325	0.0357	3.70	3.70	3.70
644788	MTT	0.9537	0.0302	3.54	22.04	5.46

The bold font is used to highlight highest scores corresponding to each dataset
Results from the pair-wise t-test indicate that MTT is significantly better than DeepViral on three datasets (2697049, 333761, and 2043570) with a p-value smaller than 0.05. On the 644788 dataset, the difference is not statistically significant

ISSN: 1471-2105