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

Area of Science:

• Synthetic Biology
• Computational Biology
• Machine Learning

Background:

• Biological network optimization is hindered by high costs, extensive labor, and limited computational tools.
• Developing efficient biological systems requires extensive experimentation and analysis.

Purpose of the Study:

• To introduce METIS, an active machine learning workflow for data-driven optimization of biological targets.
• To enable efficient optimization and prototyping of genetic and metabolic networks with minimal experimental effort.

Main Methods:

• Developed a versatile active machine learning workflow with a user-friendly online interface.
• Applied METIS to optimize cell-free transcription-translation, genetic circuits, and a synthetic CO2-fixation cycle (CETCH cycle).
• Explored vast condition spaces (10^25) using a limited number of experiments (1,000) for the CETCH cycle.

Main Results:

• Achieved one to two orders of magnitude improvement in various biological systems.
• Developed the most efficient synthetic CO2-fixation cascade to date.
• Identified key factors, interactions, and bottlenecks influencing system performance.

Conclusions:

• METIS provides a convenient and powerful approach for optimizing biological networks.
• The workflow facilitates data-driven design and rapid prototyping of synthetic biological systems.
• METIS offers customizable adjustments for diverse experimental setups and user expertise.