Leveraging large language models for metabolic engineering design.

Area of Science:

• Synthetic biology
• Metabolic engineering
• Computational biology

Background:

• Cell factory development is complex and iterative.
• Predicting effective metabolic engineering targets is challenging.
• Leveraging past successful designs is crucial for future advancements.

Purpose of the Study:

• To develop a method for extracting metabolic engineering strategies from scientific literature.
• To create a predictive model for identifying cell factory engineering targets.
• To provide a valuable resource for the metabolic engineering community.

Main Methods:

• Utilized large language models to mine research articles for metabolic engineering data.
• Constructed a comprehensive database of metabolic engineering entries, products, and organisms.
• Trained a deep learning model on the curated database for target prediction.

Main Results:

• The developed deep learning model outperformed traditional algorithms in predicting engineering targets.
• The model demonstrated strong generalization capabilities for novel products and multigene combinations.
• Experimental validation in yeast for geraniol overproduction identified several new engineering targets.

Conclusions:

• The study provides a novel dataset, chatbot, and predictive model for metabolic engineering.
• This approach efficiently leverages existing knowledge for future cell factory development.
• The findings facilitate more predictable and efficient metabolic engineering strategies.