Target-Specific Machine Learning Scoring Function Improved Structure-Based Virtual Screening Performance for...

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning

Background:

• Structure-based virtual screening (SBVS) accuracy is improved by machine learning (ML).
• Publicly available empirical data further enhances ML performance in drug discovery.
• Target-specific scoring functions may outperform universal ones in drug development.

Purpose of the Study:

• To develop a method for creating effective target-specific scoring functions.
• To apply this method to the SARS-CoV-2 3CLpro enzyme as a proof-of-concept.
• To improve the accuracy and efficiency of virtual screening in drug discovery.

Main Methods:

• Utilized BindingDB for experimental data and Smina for generating protein-ligand complexes.
• Extracted Interaction FingerPrint (IFP) and SimpleInteractionFingerPrint (SIFP) using the open drug discovery tool (oddt).
• Employed random forest classifiers and regressors with MACCS, ECFP4, ECFP6, IFP, and SIFP fingerprints.

Main Results:

• Random forest models achieved an area under the precision-recall curve of 0.80, indicating satisfactory accuracy.
• Enrichment factor analysis showed the trained scoring function outperformed Smina's generic function in molecule ranking.
• Molecular dynamics simulations confirmed the stability of top-ranked molecules in the 3CLpro active site.

Conclusions:

• The developed method provides a template for creating target-specific scoring functions against enzymes.
• This approach can significantly aid drug discovery by improving virtual screening efficiency and accuracy.
• The proof-of-concept study demonstrates the potential of ML in generating robust, enzyme-specific drug screening tools.