Fig. 3. Score function comparison for specific proteins for protein–protein docking and ligand docking.
Comparison of different score functions for different applications using the PNear metric as an indication of “funnel quality”. PNear falls between 0 (no funnel or incorrect global minimum) and 1 (the perfect funnel). The lambda parameter indicates the spread on the x-axis and is set to 4.0. Score functions are sorted from oldest to newest (left to right) and the models are colored in gray as the native (PDB) structure, then according to the score functions in order: yellow, green, cyan, and teal. A, B comparison for protein-protein docking on target with PDB ID 3eo1. The starting model is shown in dark blue—the docking partner of the starting model is too far away to be shown in the picture. The quality of the prediction improves over different score functions as indicated by tightening of the energy funnel. C, D comparison for ligand docking on target 4bqh. The native ligand pose is shown in dark blue. The quality of the prediction improves over different score functions as indicated by tightening of the energy funnel. E–H Ligand docking comparison on targets 3tll and 4uwc, respectively. Newer score functions lower the energy of an incorrect, alternative docking conformation, leading to a worse prediction.