Overview of the SAMPL6 host-guest binding affinity prediction challenge

Area of Science:

• Computational chemistry
• Drug discovery
• Supramolecular chemistry

Background:

• Accurate prediction of binding affinities is crucial for accelerating drug discovery.
• Challenges in assessing computational methods include slow dynamics and lack of high-quality data.
• Host-guest systems offer a practical approach to evaluate predictive models.

Purpose of the Study:

• To present the SAMPL6 host-guest binding affinity prediction challenge.
• To evaluate the performance of various computational methods for predicting binding affinities.
• To identify factors influencing prediction accuracy in host-guest systems.

Main Methods:

• The SAMPL6 challenge involved three hosts (octa-acid, TEMOA, CB8) and 21 guest molecules.
• Ten participating groups submitted 119 entries using diverse computational approaches.
• Methods ranged from electronic structure calculations to alchemical free energy strategies.

Main Results:

• Empirical models generally outperformed first-principle methods.
• No single method consistently achieved superior results across all host-guest systems.
• Prediction accuracy showed significant system-dependent variations, emphasizing the need for host diversity.
• Machine learning approaches using prior experimental data reduced systematic errors but not statistical correlation.

Conclusions:

• Further refinement of force field parameters is needed for improved accuracy.
• Enhanced treatment of chemical effects like buffer conditions and protonation states is essential.
• Host diversity in blind evaluations is critical for robust assessment of computational methods.