Unveiling interactions between brain areas to understand how they process information is central to Neuroscience. To achieve this, a large repertoire of connectivity estimators have been devised, comprising linear and non-linear, bivariate and multivariate ones. Their results naturally lend themselves to graph theoretical representations. The last decade has seen the rise of Graph Analysis (GA) for post-processing EEG/MEG connectivity results where concepts like small-worldness have attracted considerable attention.
Yet several network construction problems still demand consideration. The core issue is sensitivity to network construction parameters including: (i) node choice, (e.g. how to parsimoniously represent the highly correlated activity of densely packed sensors), (ii) setting the threshold for edges, or perhaps the most serious problem: (iii) the type of connectivity estimator used in network building. Namely, very popular bivariate pair-wise connectivity measures are prone to the so-called “common drive effect”, which produces spurious connections when sensors are driven by a single common source.
The advent of neuroimaging technology, especially tract-tracing findings indicating that large-scale brain networks form globally sparse hierarchical modular subnetworks with specific connection weights, challenge approaches based on binary (link/no link) undirected networks.
Currently, the methodologies for weighted networks construction are being developed. Some of them designed for BOLD time series seem especially useful and worth adaptation for EEG/MEG time series. Another current trend involves sparser networks obtained by pruning less significant connections.
Networks based on multivariate connectivity measures, considering whole set of signals into a single framework, are free of common drive effects and naturally yield sparse directed networks. They thus allow overcoming current GA undirected binary graph limitations towards explicitly incorporating weights and directionality to handle topographically specific information.
An especially promising approach for brain network construction, that complies with current anatomical and physiological evidence, is the formalism based on assortative mixing. By explicitly considering connection directionality and weight, it opens the possibility of building networks composed of anatomically segregated modules that communicate over long-range links.
We look for contributions from different backgrounds and wish to induce an open debate in a bid for new perspectives in brain networks construction. Our aim is, therefore, to stimulate discussion concerning current GA's merits and shortcomings, and search for improvements and new solutions.
We welcome original articles, opinions, and review papers focused on, but not limited to:
· Graph theoretical approaches to EEG and MEG signals
· Selection of connectivity measures suitable for building and quantifying neural networks
· Insights into the functional role of the modular structure of brain networks comprising short and long range connections
· Graph theoretical methods to capture richer and biologically more meaningful models of brain network organization
Unveiling interactions between brain areas to understand how they process information is central to Neuroscience. To achieve this, a large repertoire of connectivity estimators have been devised, comprising linear and non-linear, bivariate and multivariate ones. Their results naturally lend themselves to graph theoretical representations. The last decade has seen the rise of Graph Analysis (GA) for post-processing EEG/MEG connectivity results where concepts like small-worldness have attracted considerable attention.
Yet several network construction problems still demand consideration. The core issue is sensitivity to network construction parameters including: (i) node choice, (e.g. how to parsimoniously represent the highly correlated activity of densely packed sensors), (ii) setting the threshold for edges, or perhaps the most serious problem: (iii) the type of connectivity estimator used in network building. Namely, very popular bivariate pair-wise connectivity measures are prone to the so-called “common drive effect”, which produces spurious connections when sensors are driven by a single common source.
The advent of neuroimaging technology, especially tract-tracing findings indicating that large-scale brain networks form globally sparse hierarchical modular subnetworks with specific connection weights, challenge approaches based on binary (link/no link) undirected networks.
Currently, the methodologies for weighted networks construction are being developed. Some of them designed for BOLD time series seem especially useful and worth adaptation for EEG/MEG time series. Another current trend involves sparser networks obtained by pruning less significant connections.
Networks based on multivariate connectivity measures, considering whole set of signals into a single framework, are free of common drive effects and naturally yield sparse directed networks. They thus allow overcoming current GA undirected binary graph limitations towards explicitly incorporating weights and directionality to handle topographically specific information.
An especially promising approach for brain network construction, that complies with current anatomical and physiological evidence, is the formalism based on assortative mixing. By explicitly considering connection directionality and weight, it opens the possibility of building networks composed of anatomically segregated modules that communicate over long-range links.
We look for contributions from different backgrounds and wish to induce an open debate in a bid for new perspectives in brain networks construction. Our aim is, therefore, to stimulate discussion concerning current GA's merits and shortcomings, and search for improvements and new solutions.
We welcome original articles, opinions, and review papers focused on, but not limited to:
· Graph theoretical approaches to EEG and MEG signals
· Selection of connectivity measures suitable for building and quantifying neural networks
· Insights into the functional role of the modular structure of brain networks comprising short and long range connections
· Graph theoretical methods to capture richer and biologically more meaningful models of brain network organization
