Summary: Researchers have introduced a new tool called bimodularity that brings directionality into community detection for networks. Unlike conventional methods that cluster nodes, this approach groups edges, allowing analysts to distinguish senders from receivers and to identify paired structures called bicommunities.
When applied to systems such as traffic flows, social media, or neural activity, bimodularity exposes not only which elements belong together but also how information or influence moves between them. Tests on the C. elegans neuronal wiring diagram showed that the algorithm aligns with anatomical organization and also highlights previously unseen functional groupings, suggesting broad potential for neuroscience and other fields.
Key Facts:
- Edge-Based Clustering: Bimodularity clusters edges instead of nodes, explicitly capturing flow direction within a network.
- Bicommunities: The method reveals paired communities of senders and receivers, clarifying influence, flow, and interaction patterns.
- Proven in Neuroscience: Applied to the C. elegans connectome, bimodularity produced meaningful feedforward structures and uncovered intermediate processing pathways.
Source: EPFL
As summer winds down, many people across continental Europe are making their long journeys back north.
The return trips from the beaches of southern France, Spain, and Italy routinely congest alpine tunnels and coastal highways, creating the familiar Black Saturday bottlenecks. This annual movement is an example of a directed flow: people travel from origins to destinations, producing a network with directionality.
Network science—and specifically community detection—helps make sense of such systems. For decades, researchers have developed algorithms to detect communities: clusters of nodes that are more tightly interconnected with each other than with the rest of the network. Those tools work well for undirected networks where links are mutual, producing intuitive node-based communities that summarize shared behavior or association.
But directed networks—where relationships have a clear source and target, such as who follows whom, where traffic moves, or how signals propagate in the brain—pose additional challenges. Many established methods ignore direction or treat it inconsistently, losing important information about the flow of influence or movement.
A collaborative project between EPFL and the University of Geneva reframes the idea of a community for directed graphs by capturing both membership and flow. The resulting measure, bimodularity, provides a principled way to incorporate edge direction into community detection.
Researchers in Dimitri Van De Ville’s Laboratory of Medical Image Processing and Analysis implemented this idea in a computationally efficient algorithm. Bimodularity makes it possible to see not only which groups form together but also which groups primarily send information and which primarily receive it. “With bimodularity, we can finally distinguish senders from receivers in a network. That means finer-grained detail in how communities interact — who’s sending, and who’s receiving,” says Van De Ville.
Bimodularity enables bicommunity detection
The key innovation is edge-based clustering. Rather than grouping nodes, the algorithm groups interactions with similar directionality. By clustering edges, researchers can extract paired structures: a sending community and its corresponding receiving community. These paired communities—bicommunities—reveal the directional mappings between groups that are invisible to traditional node-based methods.
Visually, this adds a second layer to network representation. Alongside the familiar node clusters that identify who belongs together, bimodularity highlights directional clusters that show how those groups exchange information. This layered view helps answer questions such as which neighborhoods send commuters to a particular business district, which social groups tend to follow certain influencers, or how sensory signals progress through stages of neural processing.
The authors validated the framework on synthetic models and then applied it to the neuronal wiring diagram of the roundworm Caenorhabditis elegans. The method organized the network in ways that match anatomical expectations and surfaced feedforward loops and intermediate pathways in the head and body motion systems. These intermediate mappings point to possible causal chains of information flow and offer new hypotheses about network function without introducing unfounded claims.
“What’s exciting is that bimodularity doesn’t just confirm the known flow from sensory input to motion — it also reveals the intermediate steps in between, like sensory to processing and processing to motion,” says first author and PhD student Alexandre Cionca. He adds that identifying such mappings could help interpret how information travels across circuits and may inform future studies on plasticity and recovery after injury.
Beyond neuroscience, bimodularity has practical value for any field that relies on directed networks: transportation planning, epidemiology, marketing, and social media analysis, to name a few. By preserving directionality and mapping senders to receivers, the approach enhances the interpretability of community structure and opens new avenues for targeted interventions and better-informed decision making.
About this AI and neuroscience research news
Author: Michael Mitchell
Source: EPFL
Contact: Michael Mitchell – EPFL
Image: The image is credited to Neuroscience News
Original Research: Open access.
“Community detection for directed networks revisited using bimodularity” by Dimitri Van De Ville et al. PNAS
Abstract
Community detection for directed networks revisited using bimodularity
Community structure is a central feature in real-world networks. Numerous methods have been proposed to identify subsets of densely interconnected nodes using measures such as modularity. These approaches have proven effective for undirected graphs, but directed edge information has not yet been fully addressed.
In this work, the authors revisit directed communities as mappings between sending and receiving groups and formalize this concept through a measure they call bimodularity. By applying convex relaxation, bimodularity can be optimized via the singular value decomposition of a directed modularity matrix. Building on this, they propose an edge-based clustering strategy to reveal directed communities and their mappings.
The framework is demonstrated on synthetic models and then applied to the Caenorhabditis elegans connectome, where it uncovers meaningful feedforward loops in the head and body motion systems. This approach lays groundwork for a clearer understanding and reliable detection of community structures in directed networks.