Machine learning of metabolite-protein interactions from model-derived metabolic phenotypes

Area of Science:

• Systems Biology
• Metabolomics
• Proteomics

Background:

• Understanding metabolite-protein interactions is vital for elucidating how metabolism influences cellular functions.
• Current methods struggle to achieve genome-scale coverage of these interactions.
• Existing approaches often lack integration of multi-omics data.

Purpose of the Study:

• To develop and validate a computational framework for predicting metabolite-protein interactions.
• To leverage genome-scale metabolic models, fluxomics, and proteomics data for improved prediction accuracy.
• To assess the impact of feature selection and gold standard generation on classifier performance.

Main Methods:

• Trained supervised classifiers using features derived from genome-scale metabolic models.
• Integrated protein abundance and reaction flux data.
• Utilized established gold standards for metabolite-protein interactions.
• Performed comparative analysis of different features and gold standard generation methods.

Main Results:

• Features integrating fluxomic, proteomic, and metabolic phenotype data significantly improved classifier performance.
• Accurate prediction of metabolite-protein interactions was achieved in *Escherichia coli* and *Saccharomyces cerevisiae*.
• Classifier performance was robust and unaffected by the choice of gold standards for non-interacting pairs.

Conclusions:

• The developed feature set enhances the accuracy of predicting metabolite-protein interactions within metabolic contexts.
• This approach offers a scalable solution for genome-wide identification of metabolite-protein interactions.
• The study provides valuable insights into the integration of multi-omics data for systems biology research.