Science 340, 737–743 (2015)

Every chemist is well aware of the benefits of efficient catalysts. They can give significant advantages when formulating a synthetic strategy — particularly when trying to control reaction enantioselectivity — and form the basis of countless industrial processes, many of which would not be economical otherwise. However, despite their prevalence, there is still much that we do not know about complex catalytic processes. The ideal situation of rationally designing catalysts with predictable chemical reactivity is therefore an unachieved goal, and is hampered by our inability to decipher and control the way in which they work.

Led by Matthew Sigman at the University of Utah and F. Dean Toste at the University of California, Berkeley, a team of researchers have now developed a combined experimental and computational strategy to determine how the structural properties of a catalyst that interacts non-covalently with its substrate can dictate the reaction outcome. They concentrate their study on a particularly puzzling set of asymmetric C–N coupling reactions catalysed by 1,1′-bi-2-naphthol (BINOL)-derived chiral phosphoric acids, chosen specifically because the rationale for their enantioselectivity is not immediately clear. They initially generate a library of catalysts based on this BINOL–phosphoric acid scaffold with the structure of each molecule differing in a systematic way; the procedure is also repeated to create a library of systematically modified substrates. Sigman, Toste and colleagues classify each of these individual changes based on their geometric and electronic properties. Then, through experimental examination, they determine how each of these small modifications influences the course of the catalytic reaction, and in particular, the observed enantioselectivity.

A diverse network of data is generated linking specific changes in structure to reaction outcomes, and when processed using linear regression algorithms, the factors controlling the catalytic reaction and its observed enantioselectivity are revealed. The resulting computational model is then used to accurately predict the enantioselectivity of further catalyst–substrate combinations. This method may now present a general approach for the analysis of other complicated catalytic systems.