Utilizing machine learning-based QSAR model to overcome standalone consensus docking limitation in beta-lactamase...

Area of Science:

• Computational Chemistry
• Medicinal Chemistry
• Machine Learning

Background:

• Consensus docking is a standard virtual drug screening method.
• Its mathematical nature limits success rates compared to individual docking methods.
• This study addresses this limitation using machine learning.

Purpose of the Study:

• To overcome the success rate limitations of consensus docking in virtual drug screening.
• To develop and validate a machine learning-based quantitative structure-activity relationship (QSAR) model.
• To improve the identification of beta-lactamase inhibitors.

Main Methods:

• In vitro beta-lactamase inhibitory screening was performed.
• Docking protocols for AutoDock Vina and DOCK6 were optimized.
• Quantitative structure-activity relationship (QSAR) models were trained using logistic regression and random forest.
• Consensus docking results were combined with QSAR models.

Main Results:

• Optimized DOCK6 identified up to 70% of active molecules.
• Consensus analysis reduced the success rate to 50% with a 16% false positive rate.
• Random forest-based QSAR restored the success rate to 70% while maintaining a 21% false positive rate.

Conclusions:

• Random forest-based QSAR models significantly outperform logistic regression models.
• Machine learning effectively overcomes standard consensus docking limitations.
• This approach offers a more efficient strategy for beta-lactamase inhibitor discovery.