BOKEI: Bayesian optimization using knowledge of correlated torsions and expected improvement for conformer generation

Area of Science:

• Computational chemistry
• Molecular modeling
• Drug discovery

Background:

• Conformer sampling is crucial for predicting molecular properties.
• Existing methods like Bayesian Optimization Algorithm (BOA) are effective but can be improved.
• Understanding correlated torsions in small molecules is key to efficient sampling.

Purpose of the Study:

• To develop and validate a novel conformer sampling method, Bayesian Optimization with Knowledge-based Expected Improvement (BOKEI).
• To improve the efficiency of finding low-energy conformations for small molecules.
• To investigate the prevalence and nature of correlated torsions in small molecules.

Main Methods:

• Utilized bivariate von Mises distributions to model correlated torsions.
• Constrained the search space using these distributions.
• Validated BOKEI on 533 small molecules using MMFF94 and GFN2.
• Compared BOKEI against BOA with Expected Improvement (BOA-EI) and a genetic algorithm (GA).

Main Results:

• BOKEI identified lower energy conformations than BOA-EI and GA in over 60% of cases.
• Discovered correlated torsions in up to 15% of small molecules, 8x more than previously reported.
• Identified patterns reflecting steric clashes, hydrogen bonds, and π-π stacking.

Conclusions:

• BOKEI is a highly effective method for efficient conformer sampling of small molecules.
• Correlated torsions are more prevalent than previously thought and provide valuable insights.
• Improved conformer sampling will advance computational modeling applications.