An AI-driven workflow for the accelerated optimization of cell-free protein synthesis

Cell-free protein synthesis (CFPS) is a versatile tool for rapid biological prototyping. However, exploring the large number of component combinations is a very time-consuming process. Active learning (AL) is known to reduce the number of experiments required, but is rarely integrated into routine laboratory workflows. To address this, we developed a fully automated Design-Build-Test-Learn (DBTL) pipeline that streamlines this optimization process with an improved AL strategy that selects informative and diverse experimental conditions. The Design phase was created entirely using ChatGPT-4 without manual code revisions, dramatically reducing coding time. This pipeline was implemented in a modular way within the Galaxy platform, following the Findable-Accessible-Interoperable-Reusable (FAIR) principles. When applied to the optimization of colicin M and E1 in both Escherichia coli and HeLa-based CFPS systems, a 2- to 9-fold increase in yield was achieved in just four cycles. This framework enables reliable, automated workflows for routine synthetic biology.