Testing automatic methods to predict free binding energy of host-guest complexes in SAMPL7 challenge

Area of Science:

• Supramolecular Chemistry
• Computational Chemistry
• Materials Science

Background:

• Designing novel host-guest complexes is crucial for supramolecular chemistry, materials science, and biotechnology.
• Accurate prediction of binding free energy is essential for designing new host systems and identifying guest molecules.
• Existing computational methods require refinement for complex host-guest systems.

Purpose of the Study:

• To develop a computational platform for predicting host-guest complex binding free energy.
• To evaluate machine learning and physics-based approaches for this prediction task.
• To guide the development of computational pipelines through participation in the SAMPL7 challenge.

Main Methods:

• Utilized machine learning models trained on existing data.
• Employed a physics-based approach combining GFN2B functional, docking, molecular mechanics, and molecular dynamics.
• Tested and refined computational pipelines using the SAMPL7 blind challenge dataset.

Main Results:

• Machine learning predictions achieved an RMSE of 1.67 kcal/mol.
• The physics-based method achieved an RMSE of 1.45 kcal/mol, contingent on accurate binding mode identification.
• The SAMPL7 challenge provided critical insights for pipeline development.

Conclusions:

• The developed computational pipeline shows potential for predicting host-guest binding free energies.
• Further work is needed to define the scope and limitations of the machine learning models.
• Accurate binding mode identification remains a key challenge for physics-based predictions in complex systems.