MFlux

MFlux predicts central metabolic flux distributions in heterotrophic bacteria using machine learning trained on published ^13C metabolic flux analysis (13C-MFA) studies to relate environmental and genetic factors to flux changes.

Key Features:

• Dataset: Uses approximately 100 published ^13C metabolic flux analysis (13C-MFA) studies focusing on heterotrophic bacterial metabolism.
• Machine Learning Methods: Trains Support Vector Machine (SVM), k-Nearest Neighbors (k-NN), and Decision Tree models to model relationships between factors and fluxes.
• Hyperparameter Optimization: Identifies optimal parameter sets for each algorithm via systematic grid search.
• Performance Validation: Validates model performance using 10-fold cross-validation and reports SVM as the most accurate method among those tested.
• Stoichiometric Adjustment: Applies quadratic programming to adjust predicted flux profiles so they satisfy stoichiometric constraints.
• Predictive Scope: Predicts central metabolic fluxomes as functions of bacterial species, substrate type, growth rate, oxygen condition, and cultivation method.
• Data Mining: Mines existing ^13C-MFA data to uncover patterns and relationships that inform model training.

Scientific Applications:

• Microbial metabolism analysis: Predicts central metabolic pathway activity and flux distributions under varying environmental and genetic conditions.
• Inference of enzyme reaction rates and cellular function: Aids interpretation of enzyme reaction rates and overall cellular function from predicted flux patterns.
• Experimental design: Informs experimental planning by predicting flux responses to changes in conditions or genotype.
• Industrial fermentation optimization: Supports optimization of fermentation processes by predicting flux alterations under different process parameters.
• Biotechnological development: Assists development of biotechnological applications by linking environmental/genetic variables to metabolic flux outcomes.

Methodology:

MFlux mines approximately 100 published ^13C-MFA datasets, trains SVM, k-NN, and Decision Tree models with grid search for hyperparameter selection, validates performance with 10-fold cross-validation, and applies quadratic programming to enforce stoichiometric constraints on predicted flux profiles.