FitScore: a fast machine learning-based score for 3D virtual screening enrichment.

Area of Science:

• Computational Chemistry
• Cheminformatics
• Drug Discovery

Background:

• Virtual screening is crucial for drug discovery but faces challenges with large compound databases and high costs.
• Current rapid docking methods often compromise scoring accuracy, leading to poor enrichment and numerous false positives.

Purpose of the Study:

• To develop a novel scoring function, FitScore, to enhance virtual screening enrichment.
• To improve the accuracy of predicting ligand-receptor compatibility in computational drug discovery.

Main Methods:

• A two-part scoring function was developed, combining a knowledge-based component with a tunable weight matrix.
• The knowledge-based component predicts atom type probabilities within receptor environments.
• The FitScore converts these probabilities into a dimensionless score for virtual screening.

Main Results:

• The FitScore demonstrated a high degree of enrichment across standardized docking test sets.
• The scoring function effectively represents ligand-binding site compatibility.
• Improved enrichment reduces false positives in virtual screening results.

Conclusions:

• The FitScore offers a significant advancement in virtual screening enrichment for computational chemistry.
• This method addresses the need for cost-effective and accurate screening of large compound libraries.
• The FitScore is a valuable tool for accelerating drug discovery pipelines.