Boosting Pose Ranking Performance via Rescoring with MM-GBSA

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• The scoring problem in molecular docking hinders accurate identification of biologically relevant ligand poses.
• Current methods struggle to rank poses when docking scores are the primary selection criterion.

Purpose of the Study:

• To address the molecular docking scoring problem.
• To evaluate the performance of scoring functions like Molecular Mechanics/Generalized Born Surface Area (MM/GBSA) and ChemPLP.
• To introduce a novel metric, "partial matching," for assessing pose accuracy.

Main Methods:

• Self-docking simulations were performed.
• MM/GBSA and ChemPLP scoring functions were utilized.
• Heavy-atom root-mean-squared deviation (HA-RMSD) was used for comparing docked and crystallographic poses.
• A "partial matching" algorithm was developed to quantify pose accuracy.

Main Results:

• The "partial matching" algorithm effectively captures cases where ligands are largely well-docked but exhibit high HA-RMSD due to flexible side chain placement.
• Rescoring poses initially ranked by ChemPLP using MM/GBSA demonstrated improved discrimination against non-cognate-like poses.
• This rescoring approach enhanced the ranking of biologically relevant poses.

Conclusions:

• Molecular docking pose selection should consider score differences relative to the top-ranked pose, not solely the rank.
• MM/GBSA rescoring of ChemPLP poses offers a more reliable method for identifying accurate ligand-target interactions.
• The "partial matching" metric provides valuable insights into ligand sampling contributions to the scoring problem.