Lead Finder docking and virtual screening evaluation with Astex and DUD test sets

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Molecular docking is crucial for identifying drug candidates.
• Accurate scoring and sampling are essential for effective virtual screening.
• Existing software often faces limitations in solvation modeling and protein flexibility.

Purpose of the Study:

• To evaluate the performance of Lead Finder, a novel molecular docking software.
• To identify areas for improvement in docking and virtual screening algorithms.
• To assess the impact of model preparation and scoring functions on success rates.

Main Methods:

• Utilized a genetic algorithm with optimization procedures for sampling.
• Employed semi-empirical molecular mechanics functionals for scoring functions.
• Performed docking and virtual screening on benchmark datasets (Astex and DUD).
• Assessed success rates, enrichment factors, and Area Under the ROC Curve (AUC ROC).

Main Results:

• Achieved 72.9% (author models) and 74.1% (Lead Finder prepared models) docking success rates on the Astex set.
• Early enrichment factors of several tens were observed in virtual screening on the DUD set.
• AUC ROC values ranged from 0.70 to 0.74, indicating imperfect separation from the null model.
• Scoring errors due to solvation models were the primary cause of docking failures.

Conclusions:

• Lead Finder demonstrates robust performance in molecular docking and initial virtual screening.
• Improvements in solvation modeling and accounting for protein backbone flexibility are necessary for enhanced virtual screening.
• Proper ligand protonation and conformation are critical for accurate docking outcomes.