Boolean matrix logic programming for active learning of gene functions in genome-scale metabolic network models.

Area of Science:

• Computational Biology
• Systems Biology
• Machine Learning

Background:

• Genome-scale metabolic network models (GEMs) are crucial for understanding and engineering biological systems.
• Accurate prediction of genetically engineered cell behavior is often hindered by incomplete gene interaction data in GEMs.
• Learning complex genetic interactions within GEMs poses significant computational and experimental challenges.

Purpose of the Study:

• To develop a novel logic-based machine learning approach to enhance the accuracy and predictive power of GEMs.
• To guide cost-effective experimentation for biological discovery and microbial engineering.
• To create interpretable logic programs for encoding and optimizing metabolic models.

Main Methods:

• Application of logic-based machine learning methods to drive biological discovery.
• Development of Boolean Matrix Logic Programming (BMLP) for efficient evaluation of large logic programs.
• Implementation of a new system ([Formula: see text]) encoding a state-of-the-art GEM for a model bacterium.

Main Results:

• The developed system ([Formula: see text]) successfully learned gene interactions with fewer training examples compared to random experimentation.
• The approach demonstrated efficiency in overcoming the expanding experimental design space.
• The system enables rapid optimization of metabolic models for reliable biological engineering.

Conclusions:

• The novel system provides a realistic approach towards a self-driving laboratory for biological discovery.
• This facilitates microbial engineering for the production of valuable compounds.
• The interpretable logic programs enhance the reliability of metabolic model optimization.