Scribd
Upload
8 views

GraphVar Dynamic Tutorial

GraphVar 0.6 introduces several new dynamic features for analyzing time-varying connectivity matrices including: 1. Calculating dynamic summary measures such as variance, standard deviation, periodicity, and point process rate/interval across sliding windows. 2. Selecting a dynamic summary measure to perform group comparisons and correlations on. 3. Plotting mean connectivity matrices for each sliding window across subjects. 4. A new "GetComponent" feature extracts connected subnetworks from results without performing full Network Based Statistics, allowing different link thresholds to be used.

Uploaded by

Rogerio Urbano
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
8 views

GraphVar Dynamic Tutorial

GraphVar 0.6 introduces several new dynamic features for analyzing time-varying connectivity matrices including: 1. Calculating dynamic summary measures such as variance, standard deviation, periodicity, and point process rate/interval across sliding windows. 2. Selecting a dynamic summary measure to perform group comparisons and correlations on. 3. Plotting mean connectivity matrices for each sliding window across subjects. 4. A new "GetComponent" feature extracts connected subnetworks from results without performing full Network Based Statistics, allowing different link thresholds to be used.

Uploaded by

Rogerio Urbano
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 26

beta 0.

Dynamic features (and others) tutorial


Generate the Sliding Window
matrices with the „connectivity“
measure you desire:

- Pearson corr
- Partial corr
- Covariance
- Mutual inf
- etc…..

Create Connectivity matrix -> sliding windows


SW Matrices are stored (as usual) in
your workspace/results/CorrMatrix
and are loaded automatically in the
subject list

GraphVar loads SW-matrices in the subject list


If SW-matrices are loaded (also
manually via „Select Subjects Corr
Matrix“) the dynamic selection
windows will appear in the GUI

Select dynamic summary measure (graph metrics/raw matrix)


Dynamic summary
measures*:

- Variance
- Standard deviation
- Periodicity
- PointProcess: rate
- PointProcess: intervall

* definition at the end of


document

Select dynamic summary measure (graph metrics/raw matrix)


GraphVar will perform all the
operations with respect to
the dynamic summary
measure:

e.g. compute the variance of


the clustering coefficient
across the sliding windows
and export or do group
statistics on this measure

Select dynamic summary measure (graph metrics/raw matrix)


A recent paper that showed differences in
variance in Clustering Coefficient and Global
Efficiency between schizophrenia patients and
controlls (here we got the idea from)

Select dynamic summary measure (graph metrics/raw matrix)


Apply these dynamic summary measures also
to the raw matrices
(in a way similar to the „DynamicBC“ toolbox)

Select dynamic summary measure (graph metrics/raw matrix)


Example output of a group difference matrix
with respect to variance of functional
connectivity between nodes over time

Example output of a variance FC group difference matrix


If you do dynamic calculations on the raw matrix (group
comparisons or correlations) you can plot the
„DynamicSummaryMeasure“ Mean Matrix
(stored in your workspace/interimResult/… )

in this case a mean matrix of FC variance across subjects

Example output of a FC variance group difference matrix


If you do dynamic calculations on the raw matrix (group
comparisons or correlations) you can plot the
„DynamicSummaryMeasure“ Mean Matrix
(stored in your workspace/interimResult/… )

in this case a mean matrix of FC variance across subjects

Plot dynamic summary mean matrix (here variance)


If you have conducted dynamic analyses on the raw matrix,
you can plot the mean connectivity across all subjects for
each sliding window

-> „Open Single Subject“ -> Dynamic Mean Matrix

Plot mean FC for each sliding window (across subjects)


… use the „sliding window“ arrow keys to navigate through
the time points ….

Plot mean FC for each sliding window (across subjects)


If you have conducted dynamic analyses on the raw matrix,
you can plot the connectivity for each subject for each
sliding window

-> „Open Single Subject“ -> CorrMatrix

Plot sliding windows for single subjects


Select subject with arrows Sliding Window

Plot mean FC for each sliding window (across subjects)


… you can also perform Network Based Statistics with the
dynamic summary measure (e.g. variance) to identify a
network that relates to this dynamic effect

NBS with dynamic summary measures


Variance:
V = var(X) returns the variance of X for vectors. For matrices,
var(X)is a row vector containing the variance of each column of X.
For N-dimensional arrays, var operates along the first nonsingleton
dimension of X. The result V is an unbiased estimator of the
variance of the population from which X is drawn, as long as X
consists of independent, identically distributed samples.

(Matlab definition)

Dynamic summary measures


Standard Deviation:
s = std(X), where X is a vector, returns the standard deviation. The
result s is the square root of an unbiased estimator of the variance
of the population from which X is drawn, as long as X consists of
independent, identically distributed samples.

(Matlab definition)

Dynamic summary measures


Periodicity:
idea and function from http://studyforrest.org/contest_findforrestnetworks.html

“Identifying task-related activity using periodic graph properties”


Sat 01 November 2014 Lars Marstaller, Jeiran Choupan, Arend Hintze

-> Periodicity:

1. Optional: band pass filter of the time series of graph measures or connectivity
measures of the raw matrix (by default this is not done; to change see script in
src/ext/DynamicSummaryMeasures/periodicity)

2. project it into the frequency domain using a Fast Fourier transform.


3. Next, we count the number of peaks (defined as zero crossings of the first derivative)
and calculate the maximum and the median of the power spectrum.
4. From these a final singular measure is derived, which is higher the more 'peaky' the
power spectrum and hence the more periodic the changes in network properties

5. (measure = maximum/[median*peaks])

Dynamic summary measures


PointProcess – rate/interval:
idea from the paper: “Enhanced repertoire of brain dynamical states during the
psychedelic experience: Enhanced Repertoire of Brain Dynamical States. Human Brain
Mapping 35, 5442–5456. doi:10.1002/hbm.22562” by Tagliazucchi, E., Carhart-Harris, R.,
Leech, R., Nutt, D., Chialvo, D.R., 2014.

->the authors apply this to BOLD time series (in GraphVar “Raw matrix”) . We also offer to
apply this method to “graph time series” (i.e., local efficiency at different time points). It is
up to the community to determine if this approach may contribute to the evolving field of
dynamic graph theory.

-> Point Process – Rate/interval:

„ … A recent series of studies demonstrated that the continuous BOLD signal can be
transformed into a discrete point-process encoding the timings of the most functionally
relevant events [Petridou et al., 2012, Davis et al., 2013; Tagliazucchi et al., a,b, 2012a]. In
this approach, relevant events are defined by a threshold crossing (e.g. whenever the
signal departs +1 SD from its mean value).
As default in GraphVar the critical
Two interdependent observables are defined once the point-process is obtained: the threshold is set as mean + - 1 SD of the
rate (number of crossings divided by the series length) and inter-event intervals respective time series
(average temporal separation between two consecutive points).

On average, there is a clear inverse relationship between these two variables. You can change this in the scripts:
Furthermore, the rate is expected to increase (or the interval to decrease) for a signal with src/calc/pointprocess_*
a high contribution of fast frequencies.”

This approach is described as an efficient alternative to spectral analysis.

Dynamic summary measures


Do it yourself:

As always you can include your own scripts for your costum analyses

1. Add your function name in the GUI


2. Add your function name in „CalcCorrMatrix“
3. Add your function name in „CalcVars“

4. -> make sure you find all the parts where „e.g. periodicity“ is mentioned in the scrips
and copy the style of the reference functions. Also use the dimensions of the data as
we suggest (the „multidimfunc“ will do the rest)

If you have an awesome function that you would like to share with the
community through GraphVar, we would be happy to receive the
proper MATLAB code to include it in an upcoming version!

Dynamic summary measures


Additional „GetComponent“
feature in beta 0.6

Dynamic summary measures


In the previous GraphVar versions we had a button
„Show Network“

-> This old button reffered to Network Based Statistics (NBS)


… NBS was (and is) always performed with parametric derived
test-statistics (i.e., p-values) as initial link threshold

-> NBS is now performed by hitting the button „NBS“

„GetComponent“ feature
In this version we have a new feature: „GetComponent“

-> this button will NOT perform NBS but will show you the
components (i.e. subnetworks) that can be extracted from what
you see in the results viewer

-> as compared to the NBS function (which always used parametric


p-values), this feature will use the respective p-values that you
select for initial link thresholds

-> this means that you could use a link-wise controlling procedure
(e.g. random Networks/Groups FDR“) and plot the respective
subnetwork in the Network Inspector

„GetComponent“ feature
-> for „GetComponent“ there is no column in the Network Inspector telling you
the significance of this component as derived from comparison with random
data (as this is only done in NBS)

-> however, if you have applied a proper link-wise controlling procedure


(e.g. Bonferroni or FDR) you could interpret the subnetwork accordingly

„GetComponent“ feature

You might also like