A versatile active learning workflow for optimization of genetic and metabolic networks

• DOI: 10.1038/s41467-022-31245-z
• Publication Date: 2022-07-05T19:03:21Z
• Authors: Amir Pandi; Christoph Diehl; Ali Yazdizadeh Kh...; Scott A. Scholz; Elizaveta Bobkova; Léon Faure; Maren Nattermann; David Adam; Nils Chapin; Yeganeh Foroughij...; Charles Moritz; Nicole Paczia; Niña Socorro Cortina; Jean-Loup Faulon; Tobias J. Erb
• Keywords: Synthetic Biology

Optimization of biological networks is often limited by wet lab labor and cost, the lack convenient computational tools. Here, we describe METIS, a versatile active machine learning workflow with simple online interface for data-driven optimization targets minimal experiments. We demonstrate our various applications, including cell-free transcription translation, genetic circuits, 27-variable synthetic CO2-fixation cycle (CETCH cycle), improving these systems between one two orders magnitude. For CETCH cycle, explore 1025 conditions only 1,000 experiments to yield most efficient cascade described date. Beyond optimization, also quantifies relative importance individual factors performance system identifying unknown interactions bottlenecks. Overall, opens way prototyping metabolic customizable adjustments according user experience, experimental setup, laboratory facilities.