Hybrid computational workflows for reaction screening & discovery

Thijs Stuyver, Javier Emilio Alfonso-Ramos

Recent advances in machine learning (ML) and computational chemistry have opened the door to the development of hybrid computational workflows for reaction screening and discovery (Fig. 1).[1],[2],[3] In this talk, I will highlight some recent work from our research group in this field. Specifically, I will discuss how we combined automated reaction profile computation with graph neural networks to screen a chemical space of over 5M [3+2] cycloaddition reactions for bioorthogonal click potential, resulting in the identification of over 100.000 plausible click reaction of interest.[4] Furthermore, I will talk about how existing quantum chemical descriptor datasets can be repurposed to construct informative representations for downstream reactivity predictions, enabling extremely data-efficient ML-workflows.2b The latter point will be illustrated for hydrogen atom transfer reactions, where we were able to make informative predictions on datasets as small as a few hundred data points, or even smaller, with our approach. In the final part of my presentation, I will focus on our efforts on the training data generation side, by showcasing our recently developed tool for automated transition state (TS) localization, TS-tools.[6] TS-tools is a Python package that enables transition state searches at xTB or DFT level-of-theory from a reaction SMILES input. On a benchmarking dataset of mono- and bimolecular reaction pathways, TS-tools reaches an excellent success rate of 95% already at xTB level of theory. For tri-/multi-molecular pathways – typically not benchmarked when developing new TS search methods, yet common in various chemical processes, cf. solvent-, auto- and enzyme catalysis – TS- tools retains its abilities, though a DFT treatment becomes essential in many cases.

Figure 1. An overview of a typical workflow associated with machine learning accelerated computational screening.

1. B. Meyer, et al., Chem. Sci. 2018, 9, 7069-7077.
2. S. Heinen, G. Falk von Rudorff, O. A. Von Lilienfeld, J. Chem. Phys. 2021, 155, 26693.
3. N. Casetti, et al., Chem. Eur. J. 2023, 29, e202301957.
4. T. Stuyver, C. W. Coley, Chem. Eur. J. 2023, e202300387.
5. J. Alfonso-Ramos, R. Neeser, T. Stuyver, ChemRxiv. 2023, doi:10.26434/chemrxiv-2023-2n281.
6. T. Stuyver, ChemRxiv. 2024, doi:10.26434/chemrxiv-2024-st2tr.