SCOUR: a stepwise machine learning framework for predicting metabolite-dependent regulatory interactions.

Area of Science:

• Systems Biology
• Metabolic Engineering
• Computational Biology

Background:

• Metabolic network topology is conserved, but regulation diverges across species, complicating accurate modeling.
• Existing computational methods primarily focus on transcriptional regulation, with limited tools for systems-scale metabolic regulation analysis.

Purpose of the Study:

• To present a novel machine learning framework, SCOUR (stepwise classification of unknown regulation), for analyzing metabolic regulation.
• To enable the use of metabolomics and fluxomics data for inferring regulatory structures in metabolic systems.

Main Methods:

• A stepwise machine learning framework applying established algorithms to metabolic data.
• Synthetic generation of training data for identifying regulators of reaction fluxes.
• Evaluation on noiseless and noisy data with varying model sizes and topologies.

Main Results:

• SCOUR accurately identifies reaction fluxes controlled by single metabolite concentrations.
• Positive predictive values (PPVs) for identifying reactions controlled by two metabolites ranged from 6.6% to 88% depending on data conditions.
• Performance was significantly better than random classification, even under realistic conditions of low sampling frequency and high noise.

Conclusions:

• SCOUR effectively leverages metabolomics and fluxomics data to infer metabolic regulatory structures.
• The framework significantly reduces the time and effort required for experimental validation of metabolic regulatory interactions.
• SCOUR is poised to impact the development of predictive metabolic models for new organisms and pathways.