Active and machine learning-based approaches to rapidly enhance microbial chemical production

Area of Science:

• Metabolic Engineering
• Synthetic Biology
• Machine Learning Applications

Background:

• Current microbial engineering methods for renewable fuels and chemicals are limited by the need for extensive data for computational models or large-scale experimental screening.
• Developing intelligent methods is crucial for efficient strain engineering to enhance chemical production.

Purpose of the Study:

• To develop and evaluate an active and machine learning approach (ActiveOpt) for intelligent experimental guidance in microbial strain engineering.
• To minimize the number of measured datasets required to achieve optimal microbial phenotypes for chemical production.

Main Methods:

• Developed ActiveOpt, an active and machine learning framework designed to intelligently guide experimental design.
• Applied ActiveOpt to two distinct case studies involving Escherichia coli to optimize valine yields and neurosporene productivity.

Main Results:

• ActiveOpt successfully identified the best-performing microbial strains in fewer experimental iterations compared to traditional methods in both case studies.
• Demonstrated significant improvements in valine yields and neurosporene productivity through intelligent experimental design.

Conclusions:

• ActiveOpt effectively accelerates metabolic engineering by intelligently guiding experimental efforts.
• Machine and active learning approaches show great potential for rapidly achieving objectives in microbial strain development for chemical and fuel production.