FEP Augmentation as a Means to Solve Data Paucity Problems for Machine Learning in Chemical Biology

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Medicinal chemistry aims to optimize compounds for clinical trials.
• Machine learning (ML) and physics-based methods are key computational tools.
• Both ML and physics-based methods have limitations, often used independently.

Purpose of the Study:

• To overcome data scarcity in ML for drug discovery.
• To enhance ML model training using physics-based data.
• To demonstrate the synergy between FEP and ML for efficient lead optimization.

Main Methods:

• Utilized free energy perturbation (FEP) to generate virtual activity data.
• Augmented ML training datasets with FEP-generated data.
• Trained ML algorithms on the combined FEP-augmented and experimental datasets.

Main Results:

• ML models trained with FEP-augmented data showed comparable predictive accuracy to those trained on experimental data.
• Physics-based augmentation effectively addressed data paucity in ML.
• The study identified key mechanistic considerations for successful data augmentation.

Conclusions:

• The synergy of physics-based methods and ML significantly expedites lead optimization.
• FEP-augmented ML offers a powerful approach to accelerate drug discovery.
• This integrated strategy holds substantial promise for future drug development.