Abstract
Network structure or topology is the basis for understanding complex systems. Recently, higher-order structures have been considered as a new research direction that can provide new perspectives and phenomena. However, most existing studies have focused on simple 1-cycles and triangles, and few have explored the higher-order and non-trivial cycles. The current study focused on the cavity, which is a non-trivial cycle. We proposed a method to compute cavities with different orders based on pruning in the boundary matrix and calculated all cavities of the neural network of Caenorhabditis elegans (C. elegans) neural network. This study reports for the first time a complete cavity map of C. elegans neural network, developing a new method for mining higher-order structures that can be applied by researchers in neuroscience, network science and other interdisciplinary fields to explore higher-order structural markers of complex systems.
Introduction
In the field of network science, research has traditionally focused on pairwise interactions—the nodes (neurons) and edges (synapses) that form the "wiring diagram" of a system. However, biological neural networks are increasingly understood to rely on higher-order structures where groups of three or more nodes interact simultaneously. Among the most significant of these structures is the cavity.
A cavity is not merely a missing connection; it is a non-trivial, "hollow" cycle surrounded by "solid" cliques (fully connected sub-networks). In neuroscience, these cavities are hypothesized to be structural markers for specific types of information flow, allowing for the segregation and integration of signals. Despite their importance, mapping these structures has been computationally prohibitive. Previous attempts to map the C. elegans connectome were often incomplete or limited to low-order cycles. This paper presents a novel pruning algorithm that produces the first complete higher-order topological map of this model organism.
Methodology: The Pruning Algorithm
To identify these hidden structures, the researchers utilized tools from Algebraic Topology, representing the neural network as a "simplicial complex"—a multi-dimensional structure composed of points, lines, triangles, and higher-dimensional tetrahedra.
The core of the study is a three-step pruning algorithm designed to efficiently navigate the boundary matrix of the network:
- Unique Spanning Tree Generation: Standard spanning trees are not inherently unique. To ensure reproducibility and prioritize functional pathways, the authors used synapse counts (edge weights) to identify a unique "backbone" for the network.
- Boundary Matrix Pruning: The algorithm identifies "leaf cliques" (cliques that do not contribute to a closed loop) and iteratively deletes them. This pruning process isolates a set of linearly independent cycles from the broader network noise.
- Hollow Cycle Analysis: Because not every cycle is a true cavity (some may be "filled in" by higher-order cliques), the algorithm applies a final filter. This ensures that only true "hollow" cavities—those that cannot be reduced to smaller, simpler cycles—are recorded.
Dataset and Network Construction
The study analyzed the classic C. elegans neural network dataset, consisting of 297 neurons and 2,148 synapses. To facilitate topological analysis, the directed biological data was transformed into an undirected, weighted network. In instances where multiple synapses existed between two neurons, their weights were summed to provide a robust structural representation of the connection strength.
Results: The First Complete Cavity Map
The application of the pruning algorithm successfully generated the first complete map of Betti numbers ($\beta_n$) for the C. elegans connectome. Betti numbers represent the count of $n$-dimensional holes in a topological space. The findings revealed a surprisingly deep topological architecture:
- 1-cavities ($\beta_1$): 139 simple hollow loops.
- 2-cavities ($\beta_2$): 121 voids surrounded by triangles.
- 3-cavities ($\beta_3$): 4 higher-dimensional voids.
- 4-cavities and above: 0.
These results demonstrate that even a "simple" brain like that of C. elegans possesses significant higher-order structures at the second and third orders. The visualization of these cavities links specific neuron types—sensory neurons, interneurons, and motor neurons—to these topological voids, suggesting that the worm's behavioral complexity may emerge from the specific "shape" of its neural connections.
Discussion and Insights
The discovery of over 260 cavities suggests that the C. elegans brain is topologically much deeper than previously thought. These cavities represent an "architecture of silence"—specific pathways that allow the brain to process information locally without requiring global connectivity. By using weights to define the spanning tree, the authors also solved a major reproducibility problem in topological data analysis, ensuring that the most heavily connected (and thus biologically significant) pathways are prioritized.
Limitations and Future Directions
The primary limitation of this study is the treatment of the network as undirected. Biological synapses are inherently directional, and information flow is a "one-way street." Future research will focus on developing boundary matrices that account for directionality to more accurately reflect biological reality.
Additionally, the next phase of this research involves correlating these structural cavities with real-time behaviors, such as locomotion or mating, to determine if these topological markers predict functional outcomes. Finally, the efficiency of the pruning algorithm opens the door to mapping much larger and more complex connectomes, including those of Drosophila and eventually mammals.
Summary updated