Bayesian optimization alleviates the conformational bottleneck in mechanistic modeling

Thorough treatment of conformation in computational chemistry is required to capture the subtle energy differences that lead to experimental observations. Accurate quantum chemistry calculations are very expensive and evaluation of the entire ensemble found during a conformational search is often unachievable. This is especially true for complex transition state structures such as those in asymmetric catalysis. We develop a Bayesian optimization method to locate the minimum energy conformer of a chemical system, which greatly reduces the numbers of structures that must be optimized with expensive methods, such as DFT. Out of the tens or hundreds of conformers found (up to 625) for each chemical system, we find that our approach required optimization of on average approximately 15% of each system's total search space. Our method finds the lowest energy conformers of flexible transition states in asymmetric catalysis just as rapidly as for single molecules, indicating good general performance even in highly complex cases that are important to real-world chemistry.