Whole-Brain Connectomic Graph Neural Networks Enable Whole-Body Locomotion Control in Drosophila

Zehao Jin, Yanan Sui

TL;DR: A graph-based algorithm that implements the fruit fly connectome as a recurrent message passing network, enabling locomotion control of a physics-based fruit fly model.

Abstract: Whole-brain connectomics provides a structural blueprint for understanding how neural circuits generate behavior, yet its integration with embodied locomotion models remains largely unexplored. Using FlyWire, the complete connectome of an adult female Drosophila, we present fly-connectomic Graph Neural Networks (flyGNN), a graph-based algorithm that implements the fruit fly connectome as a recurrent message passing network, enabling locomotion control of a physics-based fruit fly model. Empirically, flyGNN reproduces whole-body locomotion behaviors, including gait initiation, straight walking, and turning, directly from connectome-structured dynamics. Analysis of neuron states further reveals the emergence of functional specialization during locomotion, which can be captured through low-dimensional embeddings. These results demonstrate that whole-brain structural maps can directly support embodied control, establishing a framework for investigating how connectome-derived architectures give rise to sensorimotor coordination in animals.

Length: long paper (up to 8 pages)

Domain: methods

Author List Check: The author list is correctly ordered and I understand that additions and removals will not be allowed after the abstract submission deadline.

Anonymization Check: This submission has been anonymized for double-blind review via the removal of identifying information such as names, affiliations, and URLs that point to identifying information.

Submission Number: 38