Unveiling and Steering Connectome Organization with Interpretable Latent Variables

2025-05-19

AI Summary

Bridging the gap between the structural complexity and genetic tractability of the *Drosophila* connectome remains a key challenge in understanding how low-dimensional organizational principles govern neural circuit function. Method: We propose a novel framework integrating subgraph extraction with generative modeling, leveraging FlyWire connectomic data to learn interpretable, low-dimensional circuit representations. Crucially, we embed structural semantics into a variational autoencoder (VAE), where each latent variable explicitly encodes a specific topological feature—e.g., modularity or characteristic path length—enabling semantically controllable subgraph generation. The method comprises structure-aware embedding, high-fidelity graph reconstruction, and conditional generation. Contribution/Results: Our approach enables biologically plausible, attribute-tunable synthesis of connectomic subgraphs. It provides a new paradigm for causal circuit dissection and targeted intervention by linking interpretable latent dimensions to measurable structural and functional properties.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Web data generation and simulation
• Semantics and Knowledge: Scalable techniques for the creation, curation, publication, maintenance, and consumption of large, Web-based, structured, reusable, knowledge graphs and ontologies

Abstract

The brain's intricate connectome, a blueprint for its function, presents immense complexity, yet it arises from a compact genetic code, hinting at underlying low-dimensional organizational principles. This work bridges connectomics and representation learning to uncover these principles. We propose a framework that combines subgraph extraction from the Drosophila connectome, FlyWire, with a generative model to derive interpretable low-dimensional representations of neural circuitry. Crucially, an explainability module links these latent dimensions to specific structural features, offering insights into their functional relevance. We validate our approach by demonstrating effective graph reconstruction and, significantly, the ability to manipulate these latent codes to controllably generate connectome subgraphs with predefined properties. This research offers a novel tool for understanding brain architecture and a potential avenue for designing bio-inspired artificial neural networks.

Problem

• Uncover low-dimensional principles in brain connectome organization
• Link latent dimensions to structural and functional features
• Generate connectome subgraphs with controllable predefined properties

Innovation

• Combines subgraph extraction with generative model
• Links latent dimensions to structural features
• Manipulates latent codes for controllable generation