Machine Learning Guided AQFEP: A Fast and Efficient Absolute Free Energy Perturbation Solution for Virtual Screening.

Area of Science:

• Computational chemistry
• Drug discovery and development
• Molecular modeling

Background:

• Structure-based methods are crucial in modern drug discovery.
• Virtual screening (VS) rapidly explores chemical spaces for potential drug candidates.
• Current methods like Relative Free Energy Perturbation (RFEP) are accurate but computationally expensive and require reference molecules.
• Absolute Free Energy Perturbation (AFEP) offers theoretical accuracy for hit identification but lacks throughput for VS.

Purpose of the Study:

• To develop an integrated workflow for efficient virtual screening of large chemical libraries.
• To combine active learning with a fast physics-based scoring function for improved drug discovery.
• To enable accurate binding affinity prediction without the need for reference molecules.

Main Methods:

• Developed an integrated workflow combining active learning with a fast Absolute Free Energy Perturbation (AFEP) method.
• Utilized a physics-based scoring function for enhanced accuracy in binding affinity prediction.
• Applied the workflow to screen large and diverse chemical libraries.

Main Results:

• Validated the workflow's performance in ranking structurally related ligands.
• Demonstrated significant hit rate enrichment in virtual screening.
• Showcased effective chemical space exploration through active learning.
• Reported the largest collection of free energy simulations to date.

Conclusions:

• The integrated workflow offers an efficient and accurate approach for virtual screening in drug discovery.
• The method overcomes the limitations of traditional RFEP and AFEP methods for high-throughput screening.
• This approach facilitates faster identification of promising drug candidates by combining computational efficiency with predictive power.