A graph neural network model to estimate cell-wise metabolic flux using single-cell RNA-seq data

Area of Science:

• Computational biology
• Metabolomics
• Systems biology

Background:

• Metabolic heterogeneity and cell-cell metabolic interplay are key drivers of disease treatment resistance.
• Current single-cell metabolomics technologies lack the throughput for systematic analysis of intra-tissue metabolic variations.
• A deeper understanding of cellular metabolic cooperation is needed to overcome treatment resistance.

Purpose of the Study:

• To develop a computational method for inferring cell-wise metabolic flux (fluxome) from single-cell RNA sequencing (scRNA-seq) data.
• To address the knowledge gap in intra-tissue metabolic heterogeneity and cooperative mechanisms.
• To enable downstream analyses of metabolic variations, enzyme sensitivities, and cell-cell metabolic communication.

Main Methods:

• Developed single-cell flux estimation analysis (scFEA), a novel computational method.
• Integrated a reconstructed human metabolic map as a factor graph and a probabilistic model.
• Utilized multilayer neural networks and a graph neural network-based optimization solver to link gene expression to reaction rates.
• Experimentally validated scFEA using matched scRNA-seq and metabolomics data from perturbed cellular conditions.

Main Results:

• scFEA accurately predicted metabolic flux, showing consistency with observed metabolite abundance variations.
• Analysis of public datasets revealed context- and cell group-specific metabolic variations.
• The method successfully inferred cell-wise fluxomes, enabling detailed metabolic profiling.

Conclusions:

• scFEA provides a robust computational approach to decipher single-cell metabolic heterogeneity and intercellular metabolic communication.
• This method can identify metabolic modules, sensitive enzymes, and metabolic interactions.
• scFEA advances the understanding of metabolic roles in disease, particularly in treatment resistance.