Table 2 DeepPurpose and duplex configuration model performances on BindingDB dataset

From: Improving the generalizability of protein-ligand binding predictions with AI-Bind

Model

Transductive

Semi-inductive

Inductive

 

AUROC

AUPRC

AUROC

AUPRC

AUROC

AUPRC

DeepPurpose

0.82 ± 0.004

0.48 ± 0.004

0.76 ± 0.041

0.70 ± 0.073

0.61 ± 0.074

0.43 ± 0.071

Config. Model

0.83 ± 0.011

0.50 ± 0.012

0.77 ± 0.055

0.71 ± 0.073

0.50 ± 0.00

0.30 ± 0.038

  1. DeepPurpose and the duplex configuration model perform well in both transductive and inductive tests on the benchmark BindingDB data. Both models fail to achieve good performance in the inductive test, i.e., while predicting over both unseen proteins and ligands.
Back to article page