Machine Learning and Automated Flow Synthesis for the Development of Peptide Catalysts for the Stereo- and Site-Selective Formation of C–C Bonds

PROJECT SUMMARY Control over stereo- and site- selectivity in bond formation events is of pivotal importance in organic synthesis and catalysis. There has been intense research to develop biomimetic catalysts to address these needs. One approach is the engineering of natural enzymes to accept non-native substrates or operate with non- natural reactivity. These biocatalysts have found application in various fields, including the large-scale synthesis of pharmaceuticals. However, introducing completely new-to-nature reactivity remains challenging. Another approach involves the development of small molecule peptide catalysts, which can be easily synthesized and designed to contain non-natural residues. Despite this advantage, reported small molecule peptide catalysts derive most of their structural diversity from individual residues and not the sequence. Thus, there exists a gap in the “middle ground”: peptide catalysts that have sequence-derived chemical diversity and are enriched by non-natural residues to confer the desired reactivity. The proposed research aims to leverage automated fast-flow peptide synthesis (AFPS) and machine learning to access this untapped space and develop peptide catalysts for stereo- and site-selective C–C bond- forming reactions. The AFPS platform enables precise and rapid synthesis of mid-length (~40 residues) peptide catalysts that contain non-natural residues. Machine learning will guide our exploration of the vast chemical space covered by this combinatorial peptide space. In one research direction, we will synthesize peptides that bind transition metals (Pd-OACs) and catalyze haloselective cross-coupling reactions. Neural network (NN) models will be used to correlate sequence to selectivity, inform catalyst design, and lead to iterative improvement of site selectivities. Another research direction focuses on developing peptide organocatalysts for stereoselective C–C bond formation through the Morita-Baylis-Hillman (MBH) reaction. This approach involves incorporating non-natural residues comprising thiol, phosphine, and azole functionality, and leveraging transfer learning to expand the reaction to new substrate classes. The proposed research offers an innovative blueprint for developing peptides that catalyze C–C bond formation reactions with pinpoint accuracy. It aims to pave the way for on-demand development of peptide catalysts for various reactions, leading to advances in organometallic chemistry, machine learning-guided catalyst discovery, and ultimately medicinal chemistry. The training plan and environment permits the design and study of the peptide catalysts with the Pentelute lab (MIT), Buchwald lab (MIT), and Gómez-Bombarelli lab (MIT). The proposed studies will be performed with the equipment, resources, and facilities available in these labs.