arXiv · 2025 · Preprint

Unveiling and Steering Connectome Organization with Interpretable Latent Variables

Yubin Li, Xingyu Liu, Guozhang Chen

A generative representation learning framework extracts interpretable latent variables from the fruit fly connectome to steer synthetic circuit topology.

The scientific question

Can low-dimensional latent variables uncover and steer the organization of biological connectomes?

Although whole-brain connectomes contain intricate synaptic graphs, their development from compact genetic codes points to low-dimensional organizational rules.

Encoder of a generative framework for connectome representations.
Figure 1a from the preprint ↗

The core idea

Explainable FlyWire Subgraph Encoding

Combining subgraph extraction from the Drosophila FlyWire connectome with a generative model and explainability module links compact latent dimensions to specific structural graph features.

What we found

Controllable Subgraph Generation

The learned representation accurately reconstructs biological subgraphs and enables targeted manipulation of latent variables to synthesize circuits with predefined topological properties.

Continue the question

Related reading

Explore our research

Cite this work

Yubin Li, Xingyu Liu, Guozhang Chen. Unveiling and Steering Connectome Organization with Interpretable Latent Variables. arXiv, 2025.