Source-linked AI summary
Mapping multiplex hubs in human functional brain network
Manlio De Domenico, Shuntaro Sasai, Alex Arenas
TL;DR
The paper addresses how to identify brain hubs when functional activity spans multiple frequency bands. It models each frequency component as a multiplex layer and finds that retaining all layers reveals distinct hubs and improves healthy-versus-schizophrenia discrimination over conventional networks.
Problem
Identifying brain hubs across multiple frequency bands simultaneously remains largely unexplored, despite frequency-specific networks carrying distinct centrality information.
Method
The study represents each frequency component as a multiplex-network layer and analyzes structural reducibility, centrality profiles, and group classification.
Results
Multiplex centrality profiles significantly outperform conventional profiles for distinguishing healthy controls from schizophrenic patients, while multiplex hubs differ from conventional hubs.
Takeaways & Limitations
Accurate characterization of brain functional networks requires considering information from different frequency bands simultaneously rather than aggregating or discarding it.
Takeaways & Limitations
The methodology is presented as a guideline for future studies applying multiplex measures to functional networks from variable brain states.
Abstract
from arXiv · showhide
Typical brain networks consist of many peripheral regions and a few highly central ones, i.e. hubs, playing key functional roles in cerebral inter-regional interactions. Studies have shown that networks, obtained from the analysis of specific frequency components of brain activity, present peculiar architectures with unique profiles of region centrality. However, the identification of hubs in networks built from different frequency bands simultaneously is still a challenging problem, remaining largely unexplored. Here we identify each frequency component with one layer of a multiplex network and face this challenge by exploiting the recent advances in the analysis of multiplex topologies. First, we show that each frequency band carries unique topological information, fundamental to accurately model brain functional networks. We then demonstrate that hubs in the multiplex network, in general different from those ones obtained after discarding or aggregating the measured signals as usual, provide a more accurate map of brain's most important functional regions, allowing to distinguish between healthy and schizophrenic populations better than conventional network approaches.
RESULTS
The multiplex analysis preserves frequency-specific topology, revealing distinct centrality profiles and improving discrimination between healthy and schizophrenic populations compared with conventional networks.
- Structural reducibility: No structural reduction was optimal in either group, indicating that frequency-specific functional networks carry non-redundant topological information.The quality function reached its maximum when all layers were retained.
- Structural reducibility: Layer dissimilarities differed between groups by up to 30%, with greater within-typical-band dissimilarity among healthy individuals.The comparison used pairwise quantum Jensen-Shannon distances and their signal-to-noise ratios.
- Centrality profiles: Multiplex centrality profiles showed no significant correlation with full-band or typical-band profiles, whereas the two single-layer profiles were very strongly correlated.This indicates that aggregation or frequency filtering produces centrality patterns distinct from the multiplex representation.
- Classification: Multiplex centrality profiles significantly outperformed conventional profiles in distinguishing healthy controls from schizophrenic patients, robustly across the number of selected features.The classifier varied the number of top features from 10 to 50.
- Hub characterization: Multiplex hubs formed a distinct set from conventional hubs, with group-specific and shared regions identified using the top 5% of group-averaged centrality rankings.Conventional networks showed no significant differences in their hub sets, while multiplex analysis identified distinct hubs.
DISCUSSION
Considering all frequency bands simultaneously changes the map of functional hubs and supports more informative comparisons between healthy and schizophrenic populations than conventional approaches.
- DISCUSSION: Multiplex hub characterization identifies distinct central regions in healthy controls and schizophrenic patients, whereas conventional networks show no significant hub differences.Hubs were defined as ROIs in the top 5% of group-averaged centrality; multiplex analysis produced group-specific and shared hub sets.
- DISCUSSION: The approach addresses an unresolved challenge in identifying hubs across multiple frequency bands simultaneously.Conventional resting-state analyses commonly remove frequencies below 0.01 Hz and above 0.1 Hz, while other bands may contain additional information.
- DISCUSSION: Whole-spectrum multiplex modeling captures frequency-specific information that conventional filtering or aggregation discards.The authors argue that functional-network characterization cannot rely on only selected or aggregated frequency information.
- DISCUSSION: Multiplex centrality reveals brain areas not previously classified as important for functional integration, including the absence of precuneus hubs detected by conventional views.This contrast is presented as evidence that aggregating or neglecting frequency information changes the inferred importance of regions.
- DISCUSSION: The methodology is proposed as a basis for applying additional multiplex measures to functional networks across variable brain states.The paper frames this as a guideline for future studies rather than a completed analysis of all such measures or states.
METHODS
The study constructs multiplex functional brain networks by treating frequency bands as layers, then uses multilayer centrality and classification to distinguish schizophrenia from controls.
- Network construction: The COBRE fMRI dataset is preprocessed into frequency-specific ROI connectivity networks, with nonsignificant edges removed before multiplex construction.The data include resting-state scans from schizophrenia patients and healthy controls, and connectivity is estimated across frequency bands.
- Multiplex construction: Each frequency layer is interconnected through replica nodes with parameterized weight D because the data do not determine inter-layer link strength.The study sets inter-layer links to weight D and evaluates this parameter for the classification task.
- Structural reducibility: Layer similarity is assessed using von Neumann entropy and Jensen–Shannon distance to identify redundant topological information across frequency bands.The Jensen–Shannon distance is used as a metric-compatible dissimilarity measure between layers.
- Classification: The full 12-layer multiplex network performs equal to or better than typical-band multiplex networks and classifiers that include phenotypic information.Supplementary comparisons evaluate both restricted frequency layers and the addition of phenotypic variables.
- Centrality analysis: Multiplex PageRank extends random-walk centrality across interconnected layers, and node scores are aggregated across layers into centrality profiles.The walker moves between neighboring nodes with rate 0.85 and teleports with rate 0.15; stationary probabilities provide multiplex PageRank.
- Classification: Random-forest classification uses centrality profiles, selecting approximately 30 ROIs and D = 24.7708 because this combination yields the highest classification accuracy.The selected ROI subset is obtained through exploratory ranking and a second classification round varying feature count and D.