Deep learning integration of molecular and interactome data for protein–compound interaction prediction

Table 2 Performance comparison on the unseen compound-test dataset

	AUROC	AUPRC	F-measure	Accuracy
Integrated model (molecular + network)	0.890 ± 0.039	0.842 ± 0.050	0.727 ± 0.085	0.843 ± 0.038
Single-modality model (molecular)	0.869 ± 0.027	0.786 ± 0.023*	0.657 ± 0.053	0.802 ± 0.017
Single-modality model (network)	0.831 ± 0.053	0.759 ± 0.055*	0.661 ± 0.073*	0.809 ± 0.030*
Graph CNN-based method [10]	0.804 ± 0.037*	0.679 ± 0.031*	0.637 ± 0.027	0.773 ± 0.009*
NeoDTI [13]	0.823 ± 0.067	0.773 ± 0.064*	0.621 ± 0.062*	0.805 ± 0.024*
SVM	0.765 ± 0.020*	0.603 ± 0.029*	0.689 ± 0.029	0.810 ± 0.016
Random forest	0.770 ± 0.023*	0.635 ± 0.026*	0.697 ± 0.036	0.828 ± 0.014

ISSN: 1758-2946