Source-linked AI summary
Multilayer Brain Networks
Michael Vaiana, Sarah Muldoon
TL;DR
Traditional networks are not equipped to model the brain’s multiple interactions across time, frequency, and modality. The paper frames multilayer networks through vector-labeled nodes and temporal-network examples, and concludes that retaining information alongside novel statistics has revealed previously unobserved features of brain structure and function.
Problem
Traditional networks are not equipped to model multiple brain interactions across time, frequency, or modality.
Source: §1 · p. 2Method
The paper characterizes multilayer networks through vector-labeled nodes and illustrates them with temporal networks.
Source: §2.1 · p. 5, §2.2 · p. 5Results
Combining retained information with novel network statistics has unveiled previously unobserved insights into human brain structure and function.
Source: §6 · p. 19Takeaways & Limitations
Multilayer networks are a simple, flexible data structure for quantitative analysis of complex data.
Source: §6 · p. 19Despite their utility, multilayer networks have been used in relatively few neuroscientific studies to date.
Source: §6 · p. 19
Abstract
from arXiv · showhide
The field of neuroscience is facing an unprecedented expanse in the volume and diversity of available data. Traditionally, network models have provided key insights into the structure and function of the brain. With the advent of big data in neuroscience, both more sophisticated models capable of characterizing the increasing complexity of the data and novel methods of quantitative analysis are needed. Recently multilayer networks, a mathematical extension of traditional networks, have gained increasing popularity in neuroscience due to their ability to capture the full information of multi-model, multi-scale, spatiotemporal data sets. Here, we review multilayer networks and their applications in neuroscience, showing how incorporating the multilayer framework into network neuroscience analysis has uncovered previously hidden features of brain networks. We specifically highlight the use of multilayer networks to model disease, structure-function relationships, network evolution, and link multi-scale data. Finally, we close with a discussion of promising new directions of multilayer network neuroscience research and propose a modified definition of multilayer networks designed to unite and clarify the use of the multilayer formalism in describing real-world systems.
1 Introduction
Traditional networks capture only one mode of interaction, limiting their fit to increasingly large brain datasets spanning multiple modalities, scales, and time or frequency. The introduction presents multilayer networks as a flexible extension and previews their neuroscience applications and the review’s proposed clarification of the formalism.
1 Introduction
- The growing volume of brain data motivates quantitative models that can describe interactions across modalities and scales and support meaningful analysis.The cited motivation includes retaining full frequency spectra and temporal profiles from recordings.Source: §1 · p. 2
- Traditional networks capture only a single interaction mode and are not equipped to model brain interactions across time, frequency, or modality.The introduction frames this limitation against the need to describe increasingly large, multimodal datasets.Source: §1 · p. 2
- Multilayer networks extend traditional networks to model complex interactions while retaining network simplicity and flexibility for multimodal data.The introduction describes them intuitively as interconnected networks, each representing an interaction among agents.Source: §1 · p. 2
- An fMRI–DTI example represents coherence-based functional connections and physical structural connections as separate layers linked across corresponding brain regions.The fMRI and DTI networks are each built from brain regions, then combined by connecting each region to itself across layers.Source: §1 · p. 3
- The review covers multilayer-network concepts, neuroscience applications, network statistics, future directions, and a modified definition intended to unify the field’s usage.The stated applications include disease, structure–function relationships, network evolution, and linking multiscale data.Source: §1 · p. 3
2 General Introduction to Multilayer Networks
This section introduces multilayer networks as an extension of traditional networks that represents distinct data aspects through layers and connections across layers while preserving multidimensional information for comparison. It also describes how multilayer networks can capture temporal changes in community structure, including communities being born or dying and nodes changing community affiliation.
2.1 Mathematical Definition
- A multilayer network extends a traditional network by assigning each vertex a vector label across dimensions, forming layers whose intra-layer edges encode within-aspect links and inter-layer edges connect features.Its supra-adjacency matrix represents the network and its layers as blocks, with intra-layer edges in diagonal blocks and inter-layer edges in off-diagonal blocks.Source: §2.1 · p. 4
2.2 Example: Temporal Networks
- A temporal network has node and time dimensions, with each layer representing a time point and sequential layers linked to model the flow of time.In the example, within-layer edges reflect activity similarity between brain regions, while inter-layer edges connect the same region in adjacent windows with weight ω.Source: §2.2 · pp. 5–6
2.3 Utility of Multilayer Networks
- Multilayer networks organize meaningful data subsets and link attributes across aspects, retaining multidimensional information for richer analyses; a brain-network representation improved healthy-versus-schizophrenia classification over single-layer and aggregate counterparts.Source: §2.3 · pp. 7–8
- Multilayer networks capture communities being born or dying and nodes changing community affiliation over time.These changes provide a richer description of network-structure evolution.Source: §2.3 · p. 8
3 Multilayer Networks in Neuroscience
Multilayer networks retain brain data across spatial and temporal scales, modalities, and subjects to support analyses of disease, structure–function relationships, and network change. Applications include community analysis of learning-related network change and neuron–glia models linking cellular resource distribution to synaptic stability.
3.1 Disease
- In, multilayer PageRank classified healthy versus schizophrenic individuals significantly better than single-layer counterparts, and multilayer hubs identified regions distinctive to schizophrenia.The analysis used multilayer PageRank centrality as a classifier input.Source: §3.1 · p. 9
- In, MEG multilayer networks showed reduced nodal centrality in AD versus healthy participants, unlike frequency-specific networks, and identified vulnerable hubs consistent with other imaging studies.The authors describe this as the first MEG study to identify selectively vulnerable AD hubs and suggest multilayer analysis may be essential for future MEG studies.Source: §3.1 · pp. 9–10
3.2 Structure
- In the C. elegans multilayer connectome, synaptic and extrasynaptic networks had distinct topologies but shared motifs and hubs, suggesting they can act independently and together.The layers represent synaptic, gap-junction, monoamine, and neuropeptide interactions.Source: §3.2 · p. 10
- A structural–functional multilayer network [27] showed that functional brain networks are constrained by structure and that structural connections may be necessary for positive functional correlations.Source: §3.2 · p. 11
3.3 Network Evolution
- Temporal networks retain brain-dynamics data without aggregating connections, and fMRI studies have revealed otherwise hidden features of brain dynamics during learning [16] [19].These studies parcellate the brain into regions, calculate within-layer functional similarity, and couple layers across time.Source: §3.3 · p. 11
- Maximum modularity uses a fast Louvain-like heuristic and has known problems, but can be made robust, performs well on benchmark graphs, and has worked well in multilayer practice [61, 62, 63, 64, 65, 66, 16, 18].Source: §3.3 · p. 11
- Multilayer modularity uses resolution parameters γ_s, inter-layer coupling ω, and a null matrix P_ijs; functional temporal networks typically couple each node to itself across adjacent windows with uniform ω.Because inter-layer edges are generally not measured empirically, ω becomes a model parameter.Source: §3.3 · p. 12
- Flexibility counts node community-assignment changes divided by possible changes [16]; significantly high-flexibility nodes are peripheral and significantly low-flexibility nodes core, and stronger core–periphery structure was associated with better simple motor-task learning.Source: §3.3 · p. 12
- As subjects learned a task, motor and visual systems decoupled and non-motor, non-visual systems became less integrated; cognitive control may lessen as skills become automatic.The module allegiance matrix revealed these changes across scanning sessions and subjects.Source: Figure 6 · p. 13, §3.3 · p. 14
- The motor–visual decoupling was not detectable in the fMRI signal alone, highlighting the module allegiance matrix’s role in detecting dynamic brain configurations during learning [19].Source: §3.3 · p. 14
3.4 Linking scales
- A two-layer neuron–glia network showed that glial resource distribution can provide a simple, realistic mechanism for preserving synaptic stability.Glia distribute metabolic resources to neural synapses, linking a cellular-scale process to synaptic plasticity in learning.Source: §3.4 · p. 14
4 Multilayer Network Statistics
Multilayer neuroscience has used both traditional network statistics adapted to this setting and measures developed for multilayer networks, producing novel insights. The review points to further measures and research areas that may benefit from broader use of this framework.
4 Multilayer Network Statistics
- Studies report novel insights from multilayer analysis using extended traditional statistics and measures designed specifically for multilayer networks.Examples include PageRank centrality, degree centrality [43], modularity [54], flexibility [16], core-periphery [16], and integration and recruitment [19].Source: §4 · p. 14
- Multilayer network research also spans community-structure dynamics [75] [78] [80] [82], motifs [27], and network structure [43] [84] [87].The review notes that many measures in these areas have not necessarily been used in neuroscience.Source: §4 · p. 14
- As multilayer analysis enters network neuroscience, the review argues that incorporating more of its measures into brain-network analysis would be beneficial.The cited studies motivate this direction with novel insights from multilayer analyses.Source: §4 · p. 14
5 Future Directions
The review identifies future work on multilayer network properties and dynamics, including modeling paradigms, null models, inter-layer edges, and a unifying definition. These directions concern how multilayer networks are constructed, analyzed, and represented.
5.1 Novel modeling paradigms
- Frequency-layer and time-layer paradigms have revealed more information than their single-layer counterparts, while a three-dimensional N × F × T model could combine both.The authors are unaware of a study using such a three-dimensional network; they note it could be built by combining frequency- and temporal-network methods.Source: §5.1 · p. 15
5.2 Null Models
- Develop multilayer null models tailored to neural data, potentially incorporating the brain's spatial and geometrical constraints.The review says further research is needed on such models.Source: §5.2 · pp. 15–16
5.3 Inter-layer Edges
- Few studies experimentally measure inter-layer edge strengths, and their effects on network structure and dynamics remain to be understood.Even when edge strengths are measured, their importance relative to intra-layer edges can be weighted by a global parameter; the parameter itself is not well understood.Source: §5.3 · p. 16
5.4 Unifying Definition
- The proposed general definition removes the assumption that the first label set is the nodes and treats layers as linear slices, allowing different layer choices within one network.The authors say this can better accommodate queries on realistic data and provide conceptual freedom in analyzing layers.Source: §5.4 · pp. 16–17
6 Conclusion
Multilayer networks retain information in complex brain data, and combining this retention with novel network statistics has revealed previously unobserved features of human brain structure and function. Their use in neuroscience remains limited, motivating further research and measures with improved sensitivity and specificity.
6 Conclusion
- As brain recordings become more sensitive, diverse, and ubiquitous, multilayer networks offer a flexible structure for analyzing richer data and multiple modes of interaction.Source: §6 · p. 19
- Retaining additional information with novel network statistics has revealed previously unobserved features of human brain structure and function.Source: §6 · p. 19
- Despite their utility, relatively few neuroscientific studies have incorporated multilayer networks, and the field remains in its infancy.Source: §6 · p. 19
- Future research should develop multilayer measures with improved sensitivity and specificity for potential applications.Source: §6 · p. 19
- The multilayer framework has the potential to become a prominent mode of network analysis as data become more multimodal, multitemporal, and multiscale.Source: §6 · p. 19