Home

Welcome to the comprehensive documentation for hypergraph embedding methods applied to metabolic networks!

Overview

This repository implements various embedding and kernel methods for comparing metabolic hypergraphs. The methods are designed to capture taxonomy-based features and enable comparative analysis of metabolic pathways across different organisms.

Navigation

Embedding Methods

• Bag of Hyperedges - Vector representations based on hyperedge frequencies
• Bag of Nodes - Node-based embedding methods including centrality measures

Kernel Methods

• Hypergraph Kernels - Distance and similarity measures for hypergraph comparison

Neural-based Embeddings

• Autoencoders - Neural autoencoder approaches for hypergraph representation learning
• Graph2Vec - Neural embedding approach for graph-level representations

Quick Start

1. Data Preparation: Ensure your metabolic pathway data is in the required format (see Data section for details)
2. Choose Method: Select appropriate embedding or kernel method based on your analysis needs
3. Run Analysis: Execute the corresponding script from the appropriate subdirectory
4. Results: Generated embeddings and distance matrices are saved in the data/ folder

Repository Structure

├── EmbeddingsAndKernels/
│   ├── Autoencoders/        # Neural autoencoder embeddings
│   ├── BagOfHyperedges/     # Hyperedge-based embeddings
│   ├── BagOfNodes/          # Node-based embeddings  
│   ├── Graph2Vec/           # Neural graph embeddings
│   └── Kernels/             # Distance/similarity kernels
├── data/                    # Generated embeddings and distances

Citation

If you use this code in your research, please cite our paper: "Comparing the ability of embedding methods on metabolic hypergraphs for capturing taxonomy-based features", currently under review on Algorithms for Molecular Biology and available as a preprint:

@article {Cervellini2025.07.10.663860,
	author = {Cervellini, Mattia and Sinaimeri, Blerina and Matias, Catherine and Martino, Alessio},
	title = {Comparing the ability of embedding methods on metabolic hypergraphs for capturing taxonomy-based features},
	elocation-id = {2025.07.10.663860},
	year = {2025},
	doi = {10.1101/2025.07.10.663860},
	publisher = {Cold Spring Harbor Laboratory},
	URL = {https://www.biorxiv.org/content/10.1101/2025.07.10.663860v3},
	journal = {bioRxiv}
}