mFLIP: metabolic flux interval prediction.

Area of Science:

• Systems Biology
• Metabolic Engineering
• Computational Biology

Background:

• Accurate estimation of metabolic fluxes is crucial for understanding cellular metabolism.
• Flux Variability Analysis (FVA) is a standard method for computing metabolic flux intervals but is computationally intensive for large-scale models.

Purpose of the Study:

• To develop a computationally efficient framework for predicting metabolic flux intervals.
• To provide a scalable alternative to traditional FVA for analyzing large cohorts in metabolic studies.

Main Methods:

• Proposed mFLIP, a machine learning-based framework utilizing Random Forest, XGBoost, CNN, GNN, VAE, and FT-Transformer models.
• Trained models on over 22,000 samples from metabolomics datasets (Metabolomics Workbench, MetaboLights).
• Validated model performance across six independent cancer datasets.

Main Results:

• mFLIP significantly reduced computation time for flux interval estimation from minutes to under a second.
• Random Forest and XGBoost models demonstrated superior performance with low regression errors and high classification scores.
• All proposed machine learning methods outperformed the state-of-the-art baseline in accuracy and computational efficiency.

Conclusions:

• mFLIP offers a fast and accurate alternative to FVA for metabolic flux interval estimation.
• The framework enables scalable analysis of large cohorts by leveraging supervised learning on FVA-derived data.
• mFLIP is a practical tool for large-scale metabolic studies requiring efficient flux analysis.