SARTRE

SARTRE predicts metabolite-protein interactions by integrating shadow-price features from constraint-based metabolic network analyses with supervised machine learning to identify regulatory links between metabolites and proteins.

Key Features:

• Integration of Machine Learning and Constraint-Based Modeling: Combines features derived from shadow prices, determined through flux variability analysis, with supervised machine learning to predict metabolite-protein interactions.
• Utilization of Genome-Scale Metabolic Models: Uses curated genome-scale metabolic models of Escherichia coli and Saccharomyces cerevisiae for training and validation.
• Supervised Machine Learning with Random Forest Classifiers: Implements random forest classifiers and reports average AUCs of 0.86 for Escherichia coli and 0.85 for Saccharomyces cerevisiae.
• Feature Importance Analysis: Ranks features to assess importance, highlighting shadow prices as critical predictors of interactions.
• Cross-Species Validation: Validates predictions on unseen interactions shared between Escherichia coli and Saccharomyces cerevisiae to assess transferability.

Scientific Applications:

• Understanding Metabolic Regulation: Aids elucidation of regulatory mechanisms governing protein function within metabolic networks by predicting metabolite-protein interactions.
• Comparative Analysis with Deep Learning Approaches: Provides predictions competitive with recent deep-learning methods that use diverse protein and metabolite features, enabling method comparison and benchmarking.

Methodology:

Compute shadow prices via flux variability analysis on genome-scale metabolic models of Escherichia coli and Saccharomyces cerevisiae; derive features from these shadow prices; train supervised random forest classifiers; perform feature importance ranking and cross-species validation, reporting performance as AUC.