Ricardo Peña
Speaker's Institution
Colorado State University
4:00 PM
Chemistry A101
Mixer Time
3:34 PM
Mixer Time
Chemistry B101E
Calendar (ICS) Event
Additional Information

Literature Seminar:

The discovery of new chemical reactions and the optimization of conditions, reagents, solvents, and catalysts are long-standing challenges for synthetic chemistry. From a different, more playful perspective, reaction optimization can be perceived as a game where the chemist tries to find the best combination of reagents and reaction conditions to achieve the highest yield. Since several variables influence the observed yield, many thousands of possible combinations can be tried.  To find the optimal combination in such a high-dimensional chemical space, chemists attempt to gain knowledge from mechanistic studies, although these are generally performed one reaction at a time. Alternatively, High Throughput Experimentation (HTE) can be envisaged to exhaustively sample all combinations of reactions. However, in practice, this is inefficient, costly, and unsuited to large chemical spaces.

To carry out targeted sampling of reaction parameters, Adams and Doyle have developed a Bayesian Optimization framework. This algorithmic approach iteratively suggests new reaction combinations to try from within a predefined chemical space. The Bayesian optimizer uses knowledge of high and low yielding reactions to estimate which reaction combinations the synthetic chemist should try next. These predictions are updated during the optimization campaign. Several synthetic organic reactions were subjected to Bayesian Optimization to identify high-yielding reaction conditions, catalysts, and other parameters. This approach outperforms chemists in optimizing a direct heterocycle arylation reaction in less time than human efforts. Additionally, the optimizer has demonstrated its general applicability by identifying relatively high-yielding reactions for several industrially-relevant Pd-catalyzed cross-coupling reactions.

Bayesian Optimization is introduced as a tool that avoids full-factorial sampling of a reaction space by leveraging knowledge of prior results. These predictions are iteratively improved as more knowledge is collected. This approach enables chemists to arrive more efficiently at optimal reaction combinations by reducing the number of experiments, time, and cost involved.

Shields, B. J.; Stevens, J.; Li, J.; Parasram, M.; Damani, F.; Alvarado, J. I. M.; Janey, J. M.; Adams, R. P.; Doyle, A. G. Bayesian Reaction Optimization as a Tool for Chemical Synthesis. Nature 2021, 590 (7844), 89–96. https://doi.org/10.1038/s41586-021-03213-y.