Generative Artificial Intelligence GPT-4 Accelerates Knowledge Mining and Machine Learning for Synthetic Biology.

Area of Science:

• Synthetic biology
• Biotechnology
• Computational biology

Background:

• Extracting knowledge from scientific literature for machine learning is time-consuming.
• Natural language processing (NLP) tools can automate information extraction from research articles.
• Efficient data extraction is crucial for advancing microbial engineering and biomanufacturing.

Purpose of the Study:

• To develop and validate a GPT-4 based workflow for extracting knowledge from synthetic biology publications.
• To demonstrate the utility of extracted data for machine learning-based prediction of microbial fermentation performance.
• To assess the potential of transfer learning for predicting the performance of engineered yeasts.

Main Methods:

• A GPT-4 workflow pipeline was designed using prompt engineering.
• 176 publications on *Yarrowia lipolytica* and *Rhodosporidium toruloides* were processed.
• Extracted data was structured, and feature selection was performed for machine learning model training.
• A random forest model was trained to predict fermentation titers, and transfer learning was applied.

Main Results:

• The pipeline extracted 2037 data instances from 176 publications with human intervention.
• A random forest model predicted *Yarrowia* fermentation titers with an R-squared value of 0.86 for unseen data.
• Transfer learning enabled the assessment of *R. toruloides* production potential using the trained model.

Conclusions:

• Generative AI, specifically GPT-4, can significantly accelerate knowledge extraction from scientific literature.
• Automated data extraction facilitates the development of machine learning models for predicting microbial fermentation.
• This approach holds promise for advancing biomanufacturing by streamlining data utilization and predictive modeling.