FEP Protocol Builder: Optimization of Free Energy Perturbation Protocols Using Active Learning.

Area of Science:

• Computational chemistry
• Molecular modeling
• Drug discovery

Background:

• Free energy perturbation (FEP) calculations predict binding affinity with high accuracy (∼1 kcal mol⁻¹).
• Advances in force fields, sampling algorithms, and parallel computing have driven FEP method improvements.
• Default FEP settings struggle with certain target systems, necessitating manual protocol optimization.

Purpose of the Study:

• To develop an automated workflow, FEP Protocol Builder (FEP-PB), for rapid FEP protocol generation.
• To address limitations of manual FEP protocol optimization, including time, computational cost, and potential overfitting.
• To increase the applicability of FEP calculations to a wider range of drug discovery targets.

Main Methods:

• FEP Protocol Builder (FEP-PB) employs an active-learning workflow to search the protocol parameter space.
• Iterative searching optimizes parameters to develop accurate FEP protocols.
• Validation on challenging systems like MCL1 and p97, including a real-world drug discovery setting.

Main Results:

• FEP-PB rapidly generated accurate FEP protocols for previously challenging systems (MCL1, p97).
• The automated approach required limited human intervention and outperformed expert manual optimization for the p97 system.
• Active learning provided insights into the importance of specific parameters for FEP performance.

Conclusions:

• FEP-PB is a robust tool for rapidly developing accurate FEP protocols.
• The workflow enhances the efficiency and success rate of FEP calculations in drug discovery.
• FEP-PB expands the utility of FEP technology to more complex and challenging biological targets.