Predictive methods in the field of organic synthesis have become increasingly important for medicinal and process chemists alike. One such method that has gained traction in recent years is machine learning (ML). The utilization of ML in reaction prediction relies on relatively large, well-curated datasets. While these datasets, consisting of experimental results, are becoming increasingly available, most lack the chemical diversity necessary for accurate and generalizable predictions. Alternative methods of reaction prediction exist without the need for such datasets. One of these alternative methods previously developed in the Wiest group, in collaboration with Dr. Per-Ola Norby, is quantum-guided molecular mechanics, or Q2MM, developed for stereoselectivity predictions in asymmetric catalysis reactions. Q2MM parametrizes a transition state force field resembling the stereoselecting transition state for a reaction of interest. Our in-house CatVS method virtually screens ligand-substrate combinations, calculating enantioselectivity from the predicted ΔΔG‡ of the pro-S and pro-R transition states. The Q2MM/CatVS method has shown a high correlation between calculated and experimental results (R2 ~0.9) but still computationally expensive due to exhaustive conformer searching. The automated generation of datasets for thousands of ligand-substrate combinations and their corresponding CatVS predictions was developed to train an even faster ML surrogate model. Capturing the factors responsible for stereoseletivity relies on descriptors that explain the observed Q2MM/CatVS prediction. In collaboration with the Sigman group, &gt;400 steric and electronic descriptors for bisphosphine ligands, our target for enantioselectivity predictions, have been developed using DFT single-point energy calculations. These descriptors are applied to the ligand, as well as the stereoselecting transition states in hopes of capturing the important stereoselective factors. Various model architectures are being studied to expedite stereoselectivity predictions for a reaction of interest via ML.

Q2MM/CatVS Aided machine learning for stereoselectivity predictions

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.