Please cite this toolbox as:
Kruschwitz JD, List D, Waller L, Rubinov M, Walter H, GraphVar: A user-friendly toolbox for comprehensive graph analyses of functional brain connectivity, Journal of Neuroscience Methods (2015),
New release: GraphVar 1.02: 'Turbo GLM' DOWNLOAD HERE
Important note for creating the conn matrix with partial correlations:
Minimum number of usable timepoints needed for partial correlation is at least #ROIs + 1. Alternatively you may use the SICE option (1. generate covariance matrices, 2. use SICE as threshold function)
DOWNLOAD - GraphVar:
New Release: GraphVar 1.02
Features and Highlights:
“GraphVar” is a user-friendly graphical-user-interface (GUI)-based toolbox (MATLAB) for comprehensive graph-theoretical analyses of brain connectivity, including network construction and characterization, statistical analysis on network topological measures, and interactive exploration of results. By combining together features across multiple current toolboxes, such as the Brain Connectivity Toolbox, the Graph Analysis Toolbox, and the Network Based Statistic Toolbox (BCT, Rubinov and Sporns 2010; GAT, Hosseini et al., 2012; NBS, Zalesky et al., 2010), GraphVar represents a comprehensive collection of graph analysis routines for functional neuroimaging researchers. GraphVar offers an interactive viewer that allows intuitive exploration of statistical results. Results can easily be exported and reloaded.
The program entails a detailed manual that includes usage instructions and a description of all the implemented functions, a Quick Guide for getting started, and a tutorial for getting started with sample data.
(GraphVar was developed by J.D. Kruschwitz*, D. List*, L. Waller, M. Rubinov, H. Walter and is provided under the GNU GPL v3; *equal contribution)
- GraphVar does not require MATLAB programming experience
- GraphVar contains most functions included in the Brain Connectivity Toolbox, but allows users to add custom functions which can subsequently be accessed via
- GraphVar accepts correlation matrices as input but can also generate correlation or partial correlation matrices, offers generation of connectivity matrices based on percentage bend correlations, spearman correlation and mutual information from input time series. (e.g. ROISignals_TC from the DPARSF output). Additionally GraphVar generates covariance matrices that may be used for estimating binary graphs with the sparse inverse covariance matrix option (SICE).
- Users may also provide demographic, clinical and other subject specific data (in spreadsheet format) for statistical analyses.
- GraphVar offers pipeline construction of graph networks with a choice of no, relative, absolute and significance-based thresholding, and the generation of
subject specific “null-model networks and sub-network analyses”.
- Binary and weighted network topological measures can be easily calculated, normalized, exported, and used in statistical analyses.
- Statistical analyses include general linear models (GLM) with the network measures but also on the raw connectivity matrices (i.e., network based statistics including identification of graph components).
- Statistical tests can be performed in a parametric and non-parametric fashion (i.e., testing against null-model networks, non-parametric permutation testing).
- GraphVar now supports Sliding Window analyses on the raw matrices but also with the graph topological measures!
2. Release notes:
Release info GraphVar 1.02:
1. Fixed a bug where the direction of the effect of continuous by continous
interactions was reversed in some models. This issue did not affect the
p-values of the regressors.
Release info GraphVar 1.01:
1. display mouseover non-parametric p-vals in one-dimensional graph metrics
2. enable the export of non-paramtric p-val maps from results viewer (previously only parametric p-vals were exported)
3. fixed a bug for drawing graphs in the network inspector for non-parametric p-vals<.0000001
Release info GraphVar 1.0:
1. New general linear model (GLM) framework for statistical analyses
2. Fast permutation testing via C functions
3. C++ implementation of „null_model_und_sign“ (ten times faster than the original BCT version)
4. New dynamic graph measure: Brain-Network Variability (Zhang et al. 2016)
5. Optimized results viewer
Release info GraphVar_beta_v_0.62:
1. Added two NEW "dynamic" graph measures!
- nodal flexibility and nodal promiscuity coefficient which are based on changing community assignments in an ordered multislice matrix (as in Braun et al., 2015: Dynamic reconfiguration of frontal brain networks during
executive cognition in humans)
2. Added new "regular" graph metrics:
- global cost-efficiency (as in Bassett et al.,2009): Cognitive fitness of cost-efficient brain functional networks, PNAS.
- small-world propensity (unbiased assessment of small-world structure in networks of varying densities) -> developed by Muldoon, Bridgeford and Bassett (http://arxiv.org/abs/1505.02194)
3. Finally, you can do directed analyses (graph metrics and raw matrix calculations) if you input e.g. directed granger causality matrices!
4. Added "CheckFrag": will check if network fragmentation with respect to the settings in your network construction occur
5. GraphVar now saves more output files when doing sliding window analyses (not only the dynamic summary measure as before) and will also save all results when only "calculate and Export" for further usage.
Files in your interim results folder are now (depending on what computations you do here with clustering_coeff as example):
- clustering_coef_bu_4.9_1.mat: dynamic summary measure (e.g. variance) of clustering_coef_bu across windows for each node on threshold 0.49 for all subjects
- clustering_coef_bu_4.9_1per_SW.mat: the (normalized) clustering_coef_bu for each node in each of the sliding windows on threshold 0.49 for all subjects
- clustering_coef_bu_4.9_1-rand1.mat: dynamic summary measure (e.g. variance) of clustering_coef_bu across windows for each node in the first random network on threshold 0.49 for all subjects
- clustering_coef_bu_4.9_1-rand_per_SW.mat: the clustering_coef_bu for each node in each random network in each of the sliding windows on threshold 0.49 for all subjects (i.e., cell comprised of: subjects x random networks x sliding windows)
6. Changed the normalization procedure for dynamic summary measures (this does not include "nodal flexibility/promiscuity":
- OLD normalization procedure: the dynamic summary measure of the orig. data was devided by the mean of the dynamic summary measure of the random data (there was a lot of information loss)
- NOW: first, per sliding window graph metrics are normalized as usual by division of the mean of the same graph metric derived in random networks in the same sliding window. Second, the dynamic summary measure is calculated across sliding windows of the beforehand normalized graph metrics.