Skip to main content
Fig. 1 | BMC Bioinformatics

Fig. 1

From: Improving the accuracy of high-throughput protein-protein affinity prediction may require better training data

Fig. 1

Including additional structural features improves prediction of protein-protein binding affinity. In addition to the atom-atom interaction terms evaluated in our previous study [22] we extracted additional features from protein-protein complexes in our filtered training datasets from PDBbind and the Binding Affinity Benchmark and performed cross-validation to evaluate the expected accuracy of affinity-prediction models trained using these features, when applied to new data (see Methods). We plot the Pearson correlation between predicted and experimentally-determined binding affinities for the original model (white) and the model including additional features (gray). Bars indicate standard errors. a Hydrophobicity and surface tension parameters were extracted from structural data and incorporated into the prediction model. b We calculated the root mean squared deviation (RMSD) between unbound and bound forms of the components of each protein-protein complex as well as differences in the area of each protein accessible to solvent (see Methods). These features were incorporated into prediction models. For complexes in the PDBbind database, we simulated the unbound forms by using homology modeling

Back to article page