Structure-based pose prediction: Non-cognate docking extended to macrocyclic ligands.

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Cross-docking predicts ligand binding poses for novel molecules, distinct from known cognate ligands.
• Accurate prediction is crucial for identifying new drug candidates but challenging for structurally dissimilar ligands.
• The PINC (PINC Is Not Cognate) benchmark was previously developed using temporal segregation to assess cross-docking performance.

Purpose of the Study:

• To update and extend the PINC benchmark for evaluating cross-docking performance.
• To assess the prediction accuracy for macrocyclic ligands using knowledge from non-macrocyclic ligands.
• To compare the performance of Surflex-Dock and ForceGen against other docking methods.

Main Methods:

• The PINC benchmark was expanded with 846 future ligands for ten targets, using early X-ray co-crystal structures.
• An additional set of 13 targets with 128 macrocyclic ligands was introduced, predicting poses based on non-macrocyclic ligand structures.
• Standard, automated protocols for Surflex-Dock and ForceGen were employed.

Main Results:

• Performance was comparable for both the temporally-split non-macrocyclic and macrocycle prediction sets.
• For the combined 974 ligands, Surflex-Dock and ForceGen achieved 68% success for the top-scoring pose family and 79% for the top-two.
• Correct poses were identified 92% of the time, significantly outperforming AutoDock Vina and Gnina.

Conclusions:

• The updated PINC benchmark provides a robust evaluation of cross-docking performance.
• Surflex-Dock and ForceGen demonstrate high accuracy in predicting bound poses for diverse ligands, including challenging macrocycles.
• These methods offer a significant advancement for structure-based drug discovery and virtual screening efforts.