Machine Learning-Enabled Pipeline for Large-Scale Virtual Drug Screening.

Area of Science:

• Computational chemistry
• Machine learning in drug discovery
• Bioinformatics

Background:

• Virtual screening is crucial for drug discovery but faces challenges with large compound libraries and high false positive rates.
• Efficiently filtering vast chemical libraries and distinguishing true drug candidates from decoys is a significant bottleneck.

Purpose of the Study:

• To develop and validate a machine learning-enabled pipeline for large-scale virtual screening.
• To address the challenges of library size and false positive identification in drug discovery.

Main Methods:

• Compounds were clustered by molecular properties and subjected to limited docking to reduce library size by 10-fold.
• Remaining compounds underwent individual docking screens.
• A dense neural network was trained to classify docking hits, distinguishing true from false positives.

Main Results:

• The developed pipeline effectively reduced the screening library size.
• Machine learning classification significantly improved the accuracy of identifying true positive drug candidates.
• Successful application demonstrated in screening for inhibitors against RPN11, a proteasome deubiquitinase and breast cancer target.

Conclusions:

• The machine learning pipeline offers a breakthrough for large-scale virtual screening in drug discovery.
• This approach enhances efficiency and accuracy in identifying potential therapeutic agents.
• The method holds promise for accelerating the identification of novel drug candidates for diseases like breast cancer.