(702b) Discovery of Novel Pathways Using Artificial Intelligence

In recent years, there has been rapid advancement of applying artificial intelligence to organic synthesis planning. In this work, we developed and applied a computational framework based on a template library-based approach to search for novel pathways that lead to defined molecules of interest. The foundation of the framework is automated network generation using our in-house code Pickaxe-Generic, where user-curated reaction rules are recursively applied to the input molecules and their products to create a reaction network comprised of all possible reactions and intermediates. 183 reaction rules that cover a wide range of reactions were curated by hand from organic chemistry literature. The rules are labeled and contain carefully selected auxiliary atoms to ensure the accuracy of the reaction prediction, while generalized enough to apply to all reactants with the targeted moiety. Tests were conducted against available chemical reaction datasets, including the USPTO-50k[1], and our rules reproduced the majority of the recorded reactions.

Several strategies were applied to overcome the combinatorial explosion issue during network expansion. The computational capacity of the maximum number of steps in pathways was doubled by combining the forward and retrosynthesis network. Once all possible pathways were identified, they were evaluated by selected ranking factors. Further assessments can then be carried out on the top ranked pathways to produce the final synthesis plan. Results to representative oxygenated molecules that can be used as bioprivileged molecules[2] will be discussed.

References

1. Schneider, N., N. Stiefl, and G.A. Landrum, What’s what: The (nearly) definitive guide to reaction role assignment. Journal of chemical information and modeling, 2016. 56(12): p. 2336-2346.
2. Lopez, L.M., B.H. Shanks, and L.J. Broadbelt, Identification of bioprivileged molecules: expansion of a computational approach to broader molecular space. Molecular Systems Design & Engineering, 2021. 6(6): p. 445-460.