Modular synthesis of saturated, nitrogen-containing heterocycles guided by mechanism-aware machine learning models - 1 project summary 2 successful development of new active pharmaceutical ingredients…

The National Institute of General Medical Sciences awarded The Regents of the University of California, doing business as University of California, Berkeley, $226,560 on March 1, 2026, under the Biomedical Research and Research Training program (CFDA 93.859) to develop new catalytic methods for modular synthesis of saturated, nitrogen-containing heterocycles guided by mechanism-aware machine learning models.

The research addresses a bottleneck in active pharmaceutical ingredient (API) optimization. Current synthetic methods enable modular fragment couplings to build diverse chemical structures from commercial building-block libraries, but saturated medium-sized nitrogen-containing heterocycles—among the most frequent substructures in APIs—remain underrepresented in those catalogs due to limitations in existing synthetic methods. The project will develop catalytic approaches to convert readily accessible starting materials into structurally complex heterocyclic building blocks in modular fashion, enabling more thorough and efficient API structure optimization by expanding the pool of available building blocks.

Performance occurs at Berkeley, California. The period of performance runs from March 1, 2026, through December 30, 2028. The assistance type is Project Grant.

Modular synthesis of saturated, nitrogen-containing heterocycles guided by mechanism-aware machine learning models - 1 project summary 2 successful development of new active pharmaceutical ingredients (apis) requires synthesizing many 3 structurally related compounds to optimize pharmacokinetic properties related to absorption, distribution, 4 metabolism, excretion, and toxicology. However, the expense of time and resources needed to synthesize each 5 candidate makes API optimization a bottleneck. As a result, researchers have developed a relatively short list of 6 expedient synthetic methods on which they rely to develop apis. These methods tend to involve fragment 7 couplings that forge new c-c, c-n, or c-o bonds by joining two building blocks, which allows those seeking new 8 apis to purchase libraries of suitable building blocks and explore their pairwise couplings in modular fashion. 9 the modularity of this approach means that each new fragment increases the quantity of structures that can be 10 explored in a nonlinear fashion. However, some of the most frequent substructures in apis-saturated medium- 11 sized nitrogen-containing heterocycles-are not well represented in commercial catalogues of building blocks, 12 due to limitations in state-of-the-art synthetic methods used to produce them. The ability to create custom building 13 blocks in a modular fashion, which would enable both more thorough and more efficient structure optimizations 14 of apis, would represent a significant advance in modern synthetic chemistry. However, current synthetic 15 limitations make exploration of 3D structural variants of these heterocycles difficult. 16 this proposal outlines a strategy to develop new catalytic methods that will convert readily accessible starting 17 materials into structurally complex heterocyclic building blocks in modular fashion. The modularity of the 18 proposed methods, the accessibility of starting materials, and the proposed catalyst control of stereochemistry 19 in these chiral products will make exploration of new, chiral variants of these important building blocks more 20 practical. The proposed research includes a tandem fragment coupling-cyclization approach to assemble 21 multiple building blocks to construct structurally complex, medium-sized, saturated heterocycles. Mechanistic 22 experiments and computational exploration of key mechanistic steps is proposed so that both activity and 23 selectivity of the catalysts can be understood and iteratively improved with the assistance of new data 24 representations and machine learning. Accomplishing these goals would provide practitioners access to diverse 25 structural variants of important building blocks for API development, as well as demonstrate the future role of 26 machine learning as a tool that can accelerate the development of methods with a large scope.