Synthesis of Challenging Cyclic Tetrapeptides by Machine Learning Assisted High-throughput Continuous Flow Technology

Abstract

Cyclic tetrapeptides (CTPs), which possess unique structures and diverse biological activities, are significant substances in pharmaceutical and therapeutic applications. However, the inherent ring strain in CTPs poses challenges in minimizing racemization and achieving high yields. The antiviral CTP cyclo-(Pro-Leu)₂ and the anticancer CTP cyclo-(Pro-Val)₂ were previously reported with yields of only 5% and 7%, respectively. The diverse array of peptide cyclization conditions significantly impacts the reaction outcomes, making comprehensive optimization a labor-intensive task. Herein, we integrated high-throughput continuous flow technology with machine learning to achieve rapid and comprehensive optimization for the synthesis of challenging CTPs, achieving a 5- to 7-fold improvement in yields for both cyclo-(Pro-Val)2 and cyclo-(Pro-Leu)2 compared to those reported in the literatures. Notably, with the assistance of machine learning, which achieves a root mean square error of 3.6, the optimization workload can be reduced by up to 90%. These advancements could potentially provide a solution for the rapid optimization and synthesis of valuable CTPs.