Skip to main content

Table 1 Performance comparison of three proposed models with existing methods on the baseline dataset

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

 

AUROC

AUPRC

F-measure

Accuracy

Integrated model (molecular + network)

0.972 ± 0.004

0.954 ± 0.005

0.900 ± 0.006

0.933 ± 0.004

Single-modality model (molecular)

0.956 ± 0.004*

0.927 ± 0.006*

0.868 ± 0.009*

0.911 ± 0.006*

Single-modality model (network)

0.947 ± 0.008*

0.920 ± 0.010*

0.853 ± 0.015*

0.904 ± 0.009*

Graph CNN-based method [10]

0.917 ± 0.006*

0.850 ± 0.006*

0.794 ± 0.014*

0.864 ± 0.008*

NeoDTI [13]

0.956 ± 0.005*

0.905 ± 0.016*

0.872 ± 0.006*

0.917 ± 0.004*

SVM

0.805 ± 0.009*

0.651 ±  0.012*

0.743 ± 0.012*

0.837 ± 0.006*

Random forest

0.873 ± 0.009*

0.767 ± 0.015*

0.837 ± 0.012*

0.895 ± 0.007*