Skip to main content

Table 3 Performance comparison on the hard dataset

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

 

AUROC

AUPRC

F-measure

Accuracy

Integrated model (molecular + network)

0.882 ± 0.035

0.834 ± 0.041

0.714 ± 0.064

0.836 ± 0.030

Single-modality model (molecular)

0.851 ± 0.023

0.770 ± 0.023*

0.662 ± 0.038*

0.806 ± 0.020*

Single-modality model (network)

0.780 ± 0.051*

0.706 ± 0.040*

0.601 ± 0.057*

0.784 ± 0.023*

Graph CNN-based method [10]

0.707 ± 0.038*

0.563 ± 0.083*

0.427 ± 0.132*

0.719 ± 0.043*

NeoDTI [13]

0.790 ± 0.039*

0.715 ± 0.046*

0.297 ± 0.084*

0.719 ± 0.018*

SVM

0.652 ± 0.019*

0.500 ± 0.023*

0.481 ± 0.044*

0.755 ± 0.012*

Random forest

0.605 ± 0.033*

0.452 ± 0.046*

0.364 ± 0.075*

0.728 ± 0.026*