EnzymeTuning improves enzyme-constrained metabolic modeling and proteome abundance prediction through deep learning

Area of Science:

• Systems Biology
• Metabolic Engineering
• Computational Biology

Background:

• Enzyme kinetic parameters, especially enzyme turnover numbers (kcat), are crucial for accurate enzyme-constrained genome-scale metabolic models.
• Current kinetic datasets are limited and do not fully represent in vivo enzyme activity, impacting model predictability.

Purpose of the Study:

• To develop a computational framework, EnzymeTuning, for optimizing global kcat values.
• To integrate protein degradation and synthesis rates to improve metabolic model performance.
• To assess the impact of optimized parameters on model accuracy and proteome coverage.

Main Methods:

• Developed EnzymeTuning, a generative adversarial network-based framework for kcat optimization.
• Incorporated literature-derived protein degradation constants to infer protein synthesis rates.
• Applied the framework to diverse organisms including Saccharomyces cerevisiae, Kluyveromyces lactis, Kluyveromyces marxianus, Yarrowia lipolytica, and Escherichia coli.

Main Results:

• EnzymeTuning significantly improved prediction accuracy of metabolic models.
• Expanded proteome-level coverage across various microbial species.
• Identified context-dependent enzyme utilization and adaptive resource allocation under nutrient limitation.

Conclusions:

• EnzymeTuning enhances the predictive power of genome-scale metabolic models by optimizing enzyme kinetic parameters.
• The framework provides insights into enzyme regulation and resource allocation strategies.
• EnzymeTuning holds significant potential for integrative multi-omics analyses in systems biology.