Supervised consensus scoring for docking and virtual screening.

Area of Science:

• Computational chemistry and drug discovery
• Molecular modeling and simulation

Background:

• Molecular docking is crucial for identifying novel ligands and predicting protein-ligand complex structures.
• Current scoring functions in docking often lack protein flexibility and adequate solvation treatment.
• Consensus scoring (CS) methods aim to improve performance by combining multiple scoring functions but still face limitations.

Purpose of the Study:

• To address limitations in current scoring functions by incorporating unbound ligand conformations.
• To develop a novel supervised consensus scoring (SCS) method that accounts for the protein-ligand binding process.
• To evaluate the performance of SCS against conventional CS and individual scoring functions for docking accuracy.

Main Methods:

• Developed supervised consensus scoring (SCS) utilizing unbound ligand conformations and supervised learning.
• Evaluated docking accuracy using 100 diverse protein-ligand complexes.
• Compared SCS performance against consensus scoring (CS) and eleven individual scoring functions (PLP, F-Score, LigScore, DrugScore, LUDI, X-Score, AutoDock, PMF, G-Score, ChemScore, D-score).

Main Results:

• SCS demonstrated superior docking accuracy compared to CS and individual scoring functions.
• SCS achieved success rates of 89%-91% (rmsd < 2 Å), outperforming CS (80%-85%) and others (26%-76%).
• Introduced a criterion for judging compound activity/inactivity, enhancing virtual screening utility.

Conclusions:

• Supervised consensus scoring (SCS) significantly improves docking accuracy by considering unbound ligand conformations.
• SCS offers a more robust method for virtual screening of large compound libraries.
• The SCS approach is a promising advancement for efficient drug discovery and ligand identification.