Fig. 3: Comparing DeepPurpose and the duplex configuration model.
From: Improving the generalizability of protein-ligand binding predictions with AI-Bind
a The duplex configuration model includes two layers corresponding to binding and non-binding annotations between proteins (pink nodes) and ligands (green nodes). Positive link (solid lines) and negative link (dashed lines) probabilities are determined by entropy maximization (see Methods), and used to estimate the conditional probability in transductive (Equation (7)), semi-inductive (Equation (8)), and inductive (Equation (9)) scenarios. b–d The average performance of the configuration model achieves similar results as DeepPurpose on the benchmark BindingDB data in a 5-fold cross-validation (dots represent the performance of each fold, bar height corresponds to the mean, n = 5). Breakdown of performances shows good predictive performance in transductive and semi-inductive scenarios. However, the same models have poor predictive performance in the inductive setting. Source data are provided as a Source Data file.
