Combining mechanistic and machine learning models for predictive engineering and optimization of tryptophan...

Area of Science:

• Synthetic Biology
• Metabolic Engineering
• Computational Biology

Background:

• Machine learning models are increasingly used in biological research.
• Integrating mechanistic and machine learning models can improve biological system understanding.
• Accurate genotype-to-phenotype predictions are crucial for metabolic engineering.

Purpose of the Study:

• To combine mechanistic and machine learning models for accurate genotype-to-phenotype predictions.
• To demonstrate the effectiveness of this combined approach in forward engineering metabolic pathways.
• To improve the production of aromatic amino acids in yeast.

Main Methods:

• Utilized a genome-scale model to identify engineering targets.
• Employed efficient library construction for metabolic pathway designs.
• Used high-throughput biosensor-enabled screening to train machine learning algorithms.
• Integrated mechanistic and machine learning models for predictive modeling.

Main Results:

• Achieved accurate genotype-to-phenotype predictions through the combined modeling approach.
• Successfully engineered complex aromatic amino acid metabolism in yeast within a single data-generation cycle.
• Machine learning-guided designs improved tryptophan titer by up to 74% and productivity by 43% compared to baseline designs.
• Demonstrated the efficacy of the integrated approach for directing metabolic engineering efforts.

Conclusions:

• The combination of mechanistic and machine learning models is a powerful strategy for metabolic engineering.
• This integrated approach enables efficient and accurate prediction and improvement of biological functions.
• The study provides a framework for accelerating the design-build-test-learn cycle in synthetic biology.