Abstract
Progress in scientific disciplines is accompanied by standardization of terminology. Network neuroscience, at the level of macro-scale organization of the brain, is beginning to confront the challenges associated with developing a taxonomy of its fundamental explanatory constructs. The Workgroup for HArmonized Taxonomy of NETworks (WHATNET) was formed in 2020 as an Organization for Human Brain Mapping (OHBM)-endorsed best practices committee to provide recommendations on points of consensus, identify open questions, and highlight areas of ongoing debate in the service of moving the field towards standardized reporting of network neuroscience results. The committee conducted a survey to catalog current practices in large-scale brain network nomenclature. A few well-known network names (e.g., default mode network) dominated responses to the survey, and a number of illuminating points of disagreement emerged. We summarize survey results and provide initial considerations and recommendations from the workgroup. This perspective piece includes a selective review of challenges to this enterprise, including 1) network scale, resolution, and hierarchies; 2) inter-individual variability of networks; 3) dynamics and non-stationarity of networks; 4) consideration of network affiliations of subcortical structures; and 5) consideration of multi-modal information. We close with minimal reporting guidelines for the cognitive and network neuroscience communities to adopt.
Introduction: The "Tower of Babel" in Network Neuroscience
In the last two decades, network neuroscience has transformed our understanding of brain organization. However, the field currently faces a "Tower of Babel" problem: a lack of standardized nomenclature for its most fundamental constructs. One researcher’s "Salience Network" may be another’s "Cingulo-Opercular Network." This inconsistency hinders replication, data sharing, and the synthesis of findings across different laboratories. To address this, the Workgroup for HArmonized Taxonomy of NETworks (WHATNET) was established to move the field toward a unified taxonomy.
Methods and Community Engagement
The WHATNET workgroup employed a multi-faceted approach to catalog current practices and identify technical hurdles:
- Community Survey: An online survey of 956 members of the neuroimaging community (primarily via OHBM) was conducted. Participants were presented with 93 brain network images and asked to provide a name for each.
- Technical Analysis: The group used Spatial Independent Component Analysis (sICA) and hierarchical clustering on Human Connectome Project (HCP) data (specifically the HCP_PTN820 parcellations) to demonstrate how network definitions shift across different "model orders" or levels of spatial detail.
- Literature Synthesis: The workgroup reviewed five critical challenges: spatial scale/hierarchy, inter-individual variability, time-varying dynamics, subcortical affiliations, and multi-modal convergence (e.g., fMRI vs. EEG/MEG).
Results: Consensus and the "Model Order" Trap
The survey revealed a significant consensus gap between different types of networks. There was high agreement on "unimodal" networks—those primarily involved in sensory or motor processing—and the "Default Mode Network" (DMN). Specifically, the "Somato network" achieved the highest agreement (96.97%), followed by the "Default network" (92.93%) and the "Visual network" (92.08%).
However, consensus collapsed for "higher-order" distributed networks like the Salience or Frontoparietal systems. Furthermore, images of "lure" networks—non-canonical ensembles—received 50% or less agreement, suggesting that researchers are hesitant when a pattern does not fit a well-known template.
A key insight from the technical analysis is the "Model Order" trap. Brain networks are hierarchical; at low resolution, the brain appears as a few large systems, while at high resolution, these systems fractionate into distinct subnetworks. Without a standardized taxonomy, it remains unclear whether these should be reported as parts or wholes.
Challenges to a Unified Taxonomy
The workgroup identified several persistent challenges to standardization:
- Functional vs. Anatomical Naming: Debate continues over whether to name networks by their hypothesized cognitive function (e.g., "Executive Control") or their anatomical location (e.g., "Lateral Frontoparietal").
- Individual Differences: Networks can be displaced or shaped differently across individuals, particularly in clinical populations or across the lifespan.
- Subcortical Neglect: Most current maps are "cortico-centric." A true taxonomy must incorporate the thalamus, striatum, and cerebellum, which are often excluded due to fMRI signal challenges.
- Context-Dependency: Naming a network after a single function (e.g., "The Pain Network") is often misleading, as most networks act as flexible hubs participating in diverse tasks.
Recommendations and Future Directions
To improve reporting consistency, the WHATNET workgroup proposes several minimal reporting guidelines:
- Avoid idiosyncratic cognitive labels.
- Reference findings against existing, well-defined parcellation schemes (e.g., Yeo or Gordon atlases).
- Clearly report the demographic characteristics of the sample used to define the network.
Future work will focus on "adversarial collaborations" to resolve naming conflicts between labs, cross-species mapping to find evolutionarily conserved systems, and the development of automated labeling tools to quantify spatial overlap with existing atlases. While a universal taxonomy remains a long-term goal, these steps provide a necessary foundation for more objective and reproducible network neuroscience.
Summary updated