NeuralNetworkEmbedding

Neural Network Embedding of Functional Microconnectome

This repository provides code for reproducing and extending the analyses associated with:

Shirakami, A., Hase, T., Yamaguchi, Y., & Shimono, M. (2025). Neural network embedding of functional microconnectome. Network Neuroscience, 9(1), 159–180. https://doi.org/10.1162/netn_a_00424

Overview

Neural Network Embedding (NNE) is a framework for compressing functional microconnectome matrices using a deep autoencoder and interpreting the compressed features through network metrics.

The method is designed to extract lower-dimensional representations of neuronal functional connectivity and then evaluate how these learned features relate to conventional and newly introduced network measures, including indirect-adjacent degree and neighbor hub ratio.

Key Concepts

• Neural network embedding
• Functional microconnectome
• Deep autoencoder
• Network compression
• Network metrics
• Centrality
• Indirect-adjacent degree
• Neighbor hub ratio
• Microconnectome topology
• Neural population connectivity
• Computational neuroscience
• Network neuroscience

Main Scripts

deep_autoencoder.py trains deep autoencoder models.

Usage:

python deep_autoencoder.py [depth of layers] [number of middle layer nodes]

Example:

python deep_autoencoder.py 3 5

Network_analysis.py calculates the new network metrics introduced in the paper as well as representative network metrics.

Usage:

python Network_analysis.py [directory for dataset]

Example:

python Network_analysis.py 180731001

Citation

If you use this repository in any way that contributes to a publication, preprint, thesis, presentation, software tool, benchmark comparison, dataset analysis, or derivative codebase, please cite the peer-reviewed article below.

Please cite the paper if you:

• use the original code
• modify or extend the code
• reuse part of the code in another project
• use the dataset structure or preprocessing procedure
• use the autoencoder-based network embedding framework
• use the network-metric analysis procedure
• build upon the idea of interpreting compressed microconnectome features using network metrics
• use this work as a reference for neural network embedding, functional microconnectome analysis, or network neuroscience

Peer-reviewed article

For the scientific method, results, and interpretation, please cite:

Shirakami, A., Hase, T., Yamaguchi, Y., & Shimono, M. (2025). Neural network embedding of functional microconnectome. Network Neuroscience, 9(1), 159–180. https://doi.org/10.1162/netn_a_00424

Archived software release

The GitHub repository has been archived on Zenodo:

**Shimono, M. (2026). NeuralNetworkEmbedding: Code for neural network embedding of functional microconnectome. Zenodo. https://doi.org/10.5281/zenodo.19763280

Please cite the peer-reviewed article for the scientific method, results, and interpretation, and cite the Zenodo DOI when referring specifically to this archived software release.

Suggested Citation Sentence

This repository implements neural network embedding code for compressing functional microconnectome matrices with a deep autoencoder and interpreting the learned features using network metrics.

Contact

For questions about the code, data format, or reproduction procedure, please contact the corresponding author listed in the associated paper.