Designing Chemical Reaction Arrays Using Phactor and ChatGPT

Abstract

High-throughput experimentation is a common practice in the optimization of chemical synthesis. Chemists design reaction arrays to optimize the yield of couplings between building blocks. Popular reactions used in pharmaceutical research include the amide coupling, Suzuki coupling, and Buchwald–Hartwig coupling. We show how the artificial intelligence (AI) language model ChatGPT can automatically formulate reaction arrays for these common reactions based on the literature corpus it was trained on. Critically, we showcase how ChatGPT results can be directly translated into inputs for the management software phactor, which enables automated execution and analysis of assays. This workflow is experimentally demonstrated, with modest to excellent yields of products obtained in each instance on the first attempt.

Dataset Design for Building Models of Chemical Reactivity

“…Data derived from the published literature span a wide range of substrates and reaction types, but each reactant–product combination might be reported only once or twice. In contrast, public datasets from high-throughput experimentation (HTE) exist only for a few reaction types so far (Buchwald–Hartwig amination and Suzuki coupling being the most popular datasets), although more varied datasets, both in terms of reaction types and design workflow, are emerging. , Most HTE datasets are generated through parallel plate-based chemistry in 24-, 96-, 384-, or even higher density well formats. In these experimental campaigns, some reaction variables are easy to vary via automated liquid handling capabilities (e.g., the diversity of concentrations and the combinations of additives), while other aspects (e.g., heterogeneous reactants and the diversity of solvents) are harder to vary given the practical challenges of stock solution preparation.…”

Opportunities and Challenges for Machine Learning-Assisted Enzyme Engineering

“…There is already work on developing automated evolution systems and integrating these into active learning workflows where data generated from automated experiments can train and refine ML models to suggest beneficial variants to explore further. ,, These “design-build-test-learn” cycles would enable continuous optimization of enzymes and other proteins (Figure ), as they can for small molecules . LLMs could power these automated systems, with AI flexibly adapting to perform new types of syntheses and screens with robotic scripts written on the fly. − At the same time, multiple desirable properties and activity for multiple reactions could be optimized simultaneously during protein engineering campaigns, powered by generalized ML models that can utilize multimodal representations of proteins. With ever increasing amounts of data on protein structures and sequence-fitness pairs, and new tools to conduct experiments − and make ML methods for proteins more accessible to the broader community, the future of ML-assisted protein engineering is bright.…”

The Future of ChatGPT in Medicinal Chemistry: Harnessing AI for Accelerated Drug Discovery

“…As a digital guide in this intricate domain, ChatGPT excels in route planning for intricate molecules, proposing viable pathways and intermediates to streamline the synthesis of complex compounds. [20] Furthermore, the model's expertise becomes evident in its ability to propose transformations of groups thereby outlining strategies, for synthesis improvement. The invaluable guidance extends to reagent selection, as ChatGPT skillfully recommends optimal catalysts, reagents, and conditions for specific transformations within synthetic pathways (Figure 3 shows the process by providing response to the prompt given).…”

“…The invaluable guidance extends to reagent selection, as ChatGPT skillfully recommends optimal catalysts, reagents, and conditions for specific transformations within synthetic pathways (Figure 3 shows the process by providing response to the prompt given). [20] Beyond its role as a synthesizer's companion, ChatGPT serves as an insightful tutor, unravelling the principles of retrosynthetic analysis by elucidating disconnections, bond formations, and the language of retrosynthetic arrow-pushing. By utilizing its abilities, the model effectively evaluates risks, proactively pinpointing obstacles, along retrosynthetic pathways.…”