devotg-temporal-graph-connectomics

name	devotg-temporal-graph-connectomics
description	DevoTG: Temporal Graph Neural Networks for modeling C. elegans developmental connectomics, capturing dynamic wiring through continuous-time and discrete-time dynamic graphs.
activation	developmental connectomics, temporal graph neural networks, C. elegans, connectome development, wiring dynamics
tags	["neuroscience","graph-neural-networks","temporal-networks","developmental-biology"]
arxiv_id	2606.21940
authors	["Jayadratha Gayen","Bradly Alicea"]

DevoTG: Temporal Graph Neural Networks for Developmental Connectomics

Problem

Understanding how a nervous system wires itself from birth to adulthood is fundamental in developmental neuroscience. Static graph models cannot capture the temporal dynamics of neural development.

Core Methodology

Dual Graph Representations

1. Continuous-Time Dynamic Graph (CTDG): Models cell division events from cell lineage data

   • Each division is a temporal event
   • Captures neurogenesis timing and cell fate decisions

2. Discrete-Time Dynamic Graph (DTDG): Models developing synaptic connectome

   • Spanning 8 reconstructed electron-microscopy datasets
   • Tracks 225 neurons and 858 to 2,496 connections over development

Temporal Graph Neural Network (TGN)

• Temporal memory: Maintains hidden state that evolves with graph events
• Node embeddings: Updated based on temporal neighborhood aggregation
• Link prediction: Predicts future connections based on historical dynamics

Key Results

• Lineage prediction: Mean test AUC = 0.839 ± 0.007 (5 seeds)
• Temporal advantage: Outperforms static GNN by 26 AUC points (0.577 ± 0.080)
• Connection stability classes: Identifies three classes (stable, developmental, variable)

Activation Triggers

• "developmental connectomics"
• "temporal graph neural networks"
• "C. elegans neural development"
• "wiring dynamics"
• "cell lineage prediction"
• "connectome over time"

Methodological Innovation

• Demonstrates that temporal memory is decisive for developmental prediction
• Bridges cell lineage and synaptic connectome development
• Identifies connection stability classes across development

Comparison with Static Methods

• Static GNN: 0.577 AUC (fails to capture temporal dynamics)
• Temporal GNN: 0.839 AUC (45% improvement)
• Temporal information is critical for developmental processes

Related Work

• Static Graph Neural Networks (GCN, GAT)
• Temporal Graph Networks (TGN)
• Neural development models
• C. elegans connectomics

Use Cases

• Developmental neuroscience
• Neurogenesis modeling
• Critical period identification
• Evolutionary developmental biology (evo-devo)

Reference

Gayen, J., & Alicea, B. (2026). DevoTG: Temporal Graph Neural Networks for Modeling C. elegans Developmental Connectomics. arXiv:2606.21940