A cross docking pipeline for improving pose prediction and virtual screening performance.

Area of Science:

• Computational chemistry
• Drug discovery
• Structural biology

Background:

• Molecular docking performance relies on protein structure selection.
• Using multiple protein structures accounts for flexibility but is computationally costly.
• A single protein structure may not represent all relevant conformations.

Purpose of the Study:

• To develop an efficient cross-docking pipeline for large-scale virtual screening.
• To leverage multiple target protein structures without prohibitive computational expense.
• To improve both pose prediction and virtual screening accuracy.

Main Methods:

• A cross-docking pipeline was developed for docking large molecule libraries.
• Ligand 3D shape similarity to crystallographic ligands was used for receptor selection.
• Prospective evaluation was performed using the D3R Grand Challenge 2 dataset.

Main Results:

• The proposed cross-docking pipeline achieved performance comparable or superior to single or multiple-receptor approaches.
• The method demonstrated effective pose prediction capabilities.
• Enhanced virtual screening performance was observed compared to other evaluated methods.

Conclusions:

• The novel cross-docking pipeline offers an efficient and effective strategy for virtual screening of large compound libraries.
• This approach balances computational cost with the benefits of using multiple receptor structures.
• The method shows promise for improving drug discovery pipelines.