Machine learning accelerates pharmacophore-based virtual screening of MAO inhibitors

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning

Background:

• Virtual screening is vital in drug discovery but faces limitations with large datasets and scarce data.
• Molecular docking and QSAR models struggle with speed and data dependency, respectively.

Purpose of the Study:

• To develop a universal, machine learning-based methodology for predicting docking scores.
• To accelerate virtual screening by bypassing time-consuming molecular docking procedures.

Main Methods:

• Developed a machine learning model trained on docking results.
• Utilized an ensemble approach with diverse molecular fingerprints and descriptors.
• Applied pharmacophore-constrained screening to the ZINC database.

Main Results:

• Achieved binding energy predictions 1000 times faster than classical docking.
• Identified 24 compounds from the ZINC database for synthesis and evaluation.
• Discovered weak monoamine oxidase A (MAO-A) inhibitors with promising preliminary activity.

Conclusions:

• The machine learning methodology offers a significantly faster and robust virtual screening solution.
• The approach is versatile and applicable to other biological targets without target-specific knowledge.
• This method enhances the efficiency of identifying promising drug candidates in early-stage drug discovery.