Comparative analysis of graph-based metabolic network representations using graph Kernels
Abstract
Background: Metabolism comprises a series of interconnected chemical reactions, forming a complex network that can be naturally represented as a graph. In this paper, we investigate the expressive power of three graph-based representations of metabolic networks: Abstract Metabolic Networks (AMNs), metabolic Directed Acyclic Graphs (m-DAGs), and Reaction Graphs (RGs). These representations provide a hierarchical view of metabolism, with AMNs being the most abstract, m-DAGs representing an intermediate level of detail, and RGs offering the most fine-grained view. Results: Using a comprehensive set of species from the KEGG database, we construct their metabolic networks with all three representations (i.e., AMN, m-DAG, and RG), resulting in three corresponding datasets. We apply topological analyses to highlight the distinct characteristics of the three datasets and conduct a series of metabolic network comparison tests using five selected graph kernels. The conducted experiments empirically show that, although RGs provide the highest level of detail, their expressiveness is comparable to, and in some cases slightly lower than, that of m-DAGs. AMNs, while less expressive overall, perform comparably to m-DAGs and RGs when comparing organisms from very distant taxonomic groups or those adapted to different environments, suggesting that coarse-grained representations can still capture key metabolic differences under certain conditions. Conclusions: The conducted tests empirically demonstrate the effectiveness of all three representations and the relevance of the chosen kernels for comparing metabolic networks. Additionally, these findings could serve as a useful baseline reference for exploring representation and kernel choices in metagenomic or metatranscriptomic datasets derived, e.g., from environmental samples.
Related
- LANGER: Lanthanide Affinity Network via Graph and Evolutionary Representations
- A comparative analysis of graph neural networks and classical machine learning for carbapenem-resistance prediction and genomic interpretation in Acinetobacter baumannii
- Protein–Protein Interface Residue Prediction by Integrating Protein Language Representations and Structural Graph Attention
- Explainable/interpretable artificial intelligence and knowledge graph neural network for synthetic lethality prediction: overview, challenges, and future directions
- Comments on "SpatialDG: a novel spatial domain identification method for spatially resolved transcriptomics data based on dual-graph neural network".
BibTeX
@article{garcia2026comparative,
title = {{Comparative analysis of graph-based metabolic network representations using graph Kernels}},
author = {Irene García and Pere Palmer-Rodríguez and Bessem Chouaia and Mercè Llabrés and Marta Simeoni},
journal = {Scientific Reports},
year = {2026},
month = oct,
doi = {10.1038/s41598-026-75453-9},
url = {https://www.nature.com/articles/s41598-026-75453-9},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - JOUR
TI - Comparative analysis of graph-based metabolic network representations using graph Kernels
AU - Irene García
AU - Pere Palmer-Rodríguez
AU - Bessem Chouaia
AU - Mercè Llabrés
AU - Marta Simeoni
PY - 2026
DA - 2026/10/22
JO - Scientific Reports
DO - 10.1038/s41598-026-75453-9
UR - https://www.nature.com/articles/s41598-026-75453-9
AB - Background: Metabolism comprises a series of interconnected chemical reactions, forming a complex network that can be naturally represented as a graph. In this paper, we investigate the expressive power of three graph-based representations of metabolic networks: Abstract Metabolic Networks (AMNs), metabolic Directed Acyclic Graphs (m-DAGs), and Reaction Graphs (RGs). These representations provide a hierarchical view of metabolism, with AMNs being the most abstract, m-DAGs representing an intermediate level of detail, and RGs offering the most fine-grained view. Results: Using a comprehensive set of species from the KEGG database, we construct their metabolic networks with all three representations (i.e., AMN, m-DAG, and RG), resulting in three corresponding datasets. We apply topological analyses to highlight the distinct characteristics of the three datasets and conduct a series of metabolic network comparison tests using five selected graph kernels. The conducted experiments empirically show that, although RGs provide the highest level of detail, their expressiveness is comparable to, and in some cases slightly lower than, that of m-DAGs. AMNs, while less expressive overall, perform comparably to m-DAGs and RGs when comparing organisms from very distant taxonomic groups or those adapted to different environments, suggesting that coarse-grained representations can still capture key metabolic differences under certain conditions. Conclusions: The conducted tests empirically demonstrate the effectiveness of all three representations and the relevance of the chosen kernels for comparing metabolic networks. Additionally, these findings could serve as a useful baseline reference for exploring representation and kernel choices in metagenomic or metatranscriptomic datasets derived, e.g., from environmental samples.
SN - 2045-2322
ER -