SCOUR

SCOUR predicts metabolite-dependent regulatory interactions in metabolic networks using a stepwise machine learning framework to infer regulation from metabolomics and fluxomics data.

Key Features:

• Stepwise Machine Learning Framework: A structured, multi-step classification approach that identifies unknown regulatory interactions from metabolic data.
• Generalizability Across Conditions: Validated on noiseless and noisy datasets with different sampling frequencies to assess performance under low sampling frequency and high noise.
• High Predictive Accuracy: Accurately identifies reaction fluxes controlled by the concentration of a single metabolite (primary substrate) and achieves positive predictive values of 32%–88% for two-metabolite control in noiseless data and up to 49% in noisy data.
• Synthetic Training Data Generation: Generates synthetic training datasets to enable learning regulatory relationships from metabolomics and fluxomics inputs.
• Efficiency in Experimental Validation: Ranks probable regulatory interactions to prioritize candidates for experimental validation.

Scientific Applications:

• Systems-scale metabolic regulation analysis: Identification of metabolite-dependent regulatory interactions to inform regulatory structure in metabolic network models.
• Inference from omics data: Use of metabolomics and fluxomics time-course data to predict regulators of reaction fluxes.
• Cross-organism and pathway analysis: Application to diverse organisms and pathways enabled by validation across varied noise levels and sampling frequencies.

Methodology:

SCOUR applies a stepwise machine learning classification framework trained on synthetically generated datasets derived from metabolomics and fluxomics data and evaluates performance on noiseless and noisy time-course data with varied sampling frequencies.