AIchemy’s Monthly Webinar Series – June 2026

We are delighted to welcome you to our AIchemy Hub’s monthly webinar series.

This month’s talks:

Assoc. Prof. Timonthy Cernak – University of Michigan

Talk Title: Combinatorial explosion: from atom-bond arrangements to exotic diseases

Chemical synthesis and data science are two fields that operate in synergy. Molecules and the routes to synthesize them are easily represented as graphs while automated chemical synthesis strategies allow more and more synthesis data to be captured, for instance to feed machine learning algorithms. This talk will detail our work in this area focused on a new class of amine-acid cross coupling reactions, and the computer-assisted synthesis of drugs and natural products. We have been exploring the breadth of all reactions that could exist, navigating combinatorial explosions of virtual and plausible reaction methods, routes to complex molecules, and the interconnectedness of reaction conditions, transformations, and biological functions.

Our agnostic view of reactions and their mechanisms has recently extended to diseases, with a focus on One Health. We aspire to produce medicines and treatments for health challenges in endangered species. We call this new area conservation chemistry, and examples from the frontlines of this field and lab-based research will be shared.

Simone Gallarati – University of Utah

Talk Title: Enabling data-efficient strategies for asymmetric reaction optimization and ligand discovery

In order to optimize an asymmetric reaction, machine learning (ML) models are frequently implemented to screen virtual libraries of chiral catalysts and identify candidates with superior performance. Unfortunately, such models are often poorly transferable to new reactions involving a different combination of known substrate types or an entirely unfamiliar class of compounds. In this talk, I will first introduce a descriptor generation strategy that accounts for possible changes in a reaction’s stereodetermining step with catalyst or substrate identity, allowing us to model mechanistically complex transformations involving distinct ligand and substrate types. Our ML workflow has led to the optimization of poorly performing examples reported in a substrate scope and to accurate out-of-sample predictions on unseen ligand and reaction partners.1

One limitation of inference-based ML models is the need for large virtual libraries of potential catalysts, whose curation is frequently associated with significant computational costs. In the second part of the talk, I will introduce a genetic algorithm-based pipeline2 whereby only a small population of ligands is computed and evaluated experimentally at each iteration of the optimization loop. This strategy leverages the modularity of catalyst scaffolds and is compatible with early reaction optimization campaigns, requiring the featurization and synthesis of only small batches of ligands. Overall, these workflows enable streamlined reaction development, quantitatively transferring knowledge learned on sparse data sets to novel chemical spaces.

References

(1) Gallarati, S.; Bucci, E. M.; Doyle, A. G.; Sigman, M. S. Transferable Enantioselectivity Models from Sparse Data. Nature 2026, 651, 637–646.

(2) Gallarati, S.; van Gerwen, P.; Schoepfer, A. A.; Laplaza, R.; Corminboeuf, C. Genetic Algorithms for the Discovery of Homogeneous Catalysts. CHIMIA 2023, 77 (1/2), 39.

Following the presentations, there will be time for questions from the audience.