Data from: Top-down selection of visual working memory contents is supported by alpha-band phase-synchronized oscillatory networks

Visual working memory (VWM) maintenance depends on oscillatory network
dynamics across multiple frequency bands throughout fronto-parietal and
sensory brain areas. However, whether these networks reflect the active
maintenance of visual information content or serve top-down control
processes has remained unresolved. To address this, we used concurrent
magneto- and electroencephalography (M/EEG) to measure brain activity
during VWM tasks, in which the memory content was parametrically
controlled. Using new edge-level analysis for source-connectivity
networks, we disentangled connections and subnetworks underlying the
maintenance of specific contents from those supporting feature-general
VWM. We show here that long-range high-alpha band (α, 11– 13 Hz) phase-
synchronization networks carry out a dual role in these VWM functions.
α-band subgraphs localized to the visual areas are feature-selective and
maintain the contents of VWM. In contrast, the high α-band subgraph in the
fronto-parietal areas was shared across memory contents, suggesting that
it forms the content-agnostic executive core of VWM. We propose that α-
band synchronization across distinct, but yet interconnected, subgraphs
support the active maintenance of feature representations and their
top-down selection.  

Recording and processing of this data is described in the
article.
Haque H, et al.: "Top-down selection
of visual working memory contents is supported by alpha-band
phase-synchronized oscillatory networks. Imaging
Neuroscience."

Data archive for: **Top-down selection of visual working memory contents
is supported by alpha-band phase-synchronized oscillatory networks**
Authors: Hamed Haque, Sheng H. Wang, Felix Siebenhühner, Edwin M.
Robertson, J. Matias Palva, Satu Palva Year: 2025 Contact: Hamed Haque,
[[email protected]](mailto:[email protected]);
[[email protected]](mailto:[email protected]) This data
archive contains synchronization data of twenty healthy human participants
performing a visual working memory (VWM) task while being scanned using
concurrent electroencephalography (EEG) and magnetoencephalography (MEG).
For each analysis, the relevant input and supporting files that allow
replication of the workflow and results are included. Scripts used for
these analysis can be found in GitHub
([https://github.com/palvalab/vwm_synchronization]()). In the present work
we analyse large-scale synchronization networks in multiple frequency
bands during a VWM task. We identified distinct inter-areal networks in
the alpha and theta-bands, with the alpha-band synchronization network
shown to be localized to feature-selective visual areas. We also
identified subgraphs in the alpha-band that were shared across memory
contents and thus formed the executive core of VWM. ## **Description of
the Data and file structure** ### **Summary of data** Data in this
repository are structured in 22 different folders. The metadata folder
contain various supporting files necessary for proper analysis and
interpretation of the rest of the dataset. The plot_data folder contains
various intermediary and output files that are necessary for performing
each of the analysis. The rest of the 20 folders correspond to each of the
20 subjects of the study. They are structured in the format S0** with the
last two digits corresponding to the unique identifier for each subject.
### **Usage notes** All files in the dataset included in the repository
fall into one of the three file formats: **.csv** A CSV (comma-separated
values) file is a plain text file used to store tabular data, with each
row representing values separated by a comma. CSV files can be opened with
a standard text editor such as Notepad, Microsoft Excel, or imported into
Python using the csv or pandas libraries with the csv.reader() or
pandas.read_csv() functions respectively. **.xlsx** These are zipped,
XML-based excel files created by Microsoft Excel also used to store
tabular data. XLSX files can be opened and edited using Microsoft Excel or
imported into Python using the pandas.read_excel() function. **.tdms** A
Technical Data Management Streaming (TDMS) is a file format designed by
National Instruments (NI) to store measurement data. Data are stored in a
clear hierarchical structural organized by file, group, and channel. TDMS
files can be opened directly using LabVIEW, in Microsoft Excel using the
‘TDM Excel Add-In for Microsoft Excel’ tool, or in Python using the nptdms
library through the nptdms.TdmsFile function. Routines for reading TDMS
files in Python are provided in the GitHub repository associated with this
dataset. Below is example code for viewing TDMS files in python: ``` from
nptdms import TdmsFile tdms_file =
TdmsFile.read("your_file.tdms") data = tdms_file.as_dataframe()
for group in tdms_file.groups(): for channel in group.channels():
print(f"{group.name}/{channel.name}: {channel.data}") ``` ##
**Metadata** The folder metadata contain various supporting files that
typically span several different analysis. **__group_statistics_csv** The
folders within ‘__group_statistics_csv’ contain synchronization matrices
at the group level for different conditions. These are group level
connectomes for the Pearson correlation between phase synchronization and
behavioural performance. The folders are arranged in the format:
3x2_Feature_x_Obj EdgeD  Pearson*conditiontail.* The *condition* refer to
the feature and number of objects to be memorized during the VWM while
*tail* refers the statistics performed at either the positive or negative
tail. Within each of these subfolders, the files are stored in the format:
Phase-Phase 1-1 Original No-Surrogates cPLV Lag_1-0 Lag=1.000 Low =
*frequency* hi = *frequency* parc2009_200AFS *measure*.csv. The
*frequency* refers to the frequency in which the source-reconstructed MEEG
data was filtered to before the synchronization was estimated. The
*measure* refers to different ways of presenting the significant matrix:
for each edge, ES is the effect size, p-value is the p-value, sign-stat is
the r statistic value of the Pearson correlation test. In each of these
csv files, the adjacency matrices are in the shape 200 x 200 with the
three time windows stacked vertically, resulting in a final shape of 600 x
200. **difference_masks** These contain the group level statistical mask
for edges where the contrast difference between visual features was
significant. There are two files, Alpha.csv and Theta.csv and they both
contain 200 x 200 binary matrices. A 1 indicates a significant edge while
a 0 indicates a non-significant edge. **Morphing operators** This folder
contains, for each subject, the morphing matrix that can convert the
400-parcels into the 200 parcels of the split-Destrieux parcellation. The
morphing operator is of the shape 400 x 200, with each column representing
each parcel of the 200 parcel atlas and each row the contribution of each
parcel of the 400 parcel atlas. **behavioral_df.csv** Provides the Hit
Rates for each subject (S001 to S020), per condition (Shape, Color, or
Spatial) and load (2 or 4). **cf_matrix_high_full.csv** The
cross-frequency (CF) matrix used for computing inter-areal *n:m*
synchronization, with *n* representing the low frequency and *m* the high
frequency. Frequencies used as low frequencies (first column) and as high
frequencies (2nd to 9th column) for CF ratios 1:2—1:9. All frequencies
appear in the first column. **current_DEM.csv** Contains the Deny Edge
Matrix (DEM) which is a binary 200 x 200 matrix. 1’s indicate the edge
should be included for analysis while 0’s indicate that the edge should be
excluded from group analysis. The DEM is meant to exclude poorly
reconstructable parcel edges (based on parcel fidelity and cross-patch
PLV) from group level analyses. **Functional_boundaries.csv** Label of
each parcel indicating which functional visual subsystem it belongs to.
The file contains a single column of length 200 with each row representing
a parcel. The value corresponds to the identifier of specific visual
subsystems. **Patch_Grouping_divisions.csv** Label of each parcel
indicating which of the 7 Yeo subsystems it belongs to. Each of the 200
rows corresponds to a parcel with the first column indicating the Yeo
subsystem the parcel belongs to. ## **Plot_data** The folder ‘plot_data’
contains the files specific to each of the main analyses performed in the
study. ### **Figure_2** **Contrast_betn_features:** Contains the K values
(fraction of significant edges) for each frequency for each contrast
difference between visual features. Each file in this folder represents a
contrast between two conditions, with the positive and negative tails in
separate files. The first column is the frequency and the second column
the K value. **Graph_strength:** The two files in this folder,
GS_Alpha.csv and GS_Theta.csv, contain the mean graph strength for
retention of each visual feature. The first column has the visual feature
(Shape, Color, Spatial) and the second column the mean graph strength,
which is the average iPLV value of the significant inter-areal network.
Each row represents a subject. **Wilcoxon:** Same as
‘Contrast_betn_features’ but rather than presenting the K values for
differences between conditions, contains the K values for the significant
edges after a one-sample Wilcoxon signed-rank test for each visual
feature. The ‘jackknife’ folder contains the same contrasts as the parent
folder but presents the K values for each jackknife resample. ###
**Figure_3** **Single_condition:** Contains the group-level connectomes
for the Wilcoxon signed-rank test between alpha phase synchronization of
the retention and baseline period. Each file contains a 200 x 200
adjacency matrix with the non-zero values being the average iPLV value
when the edge is significant. **Subsystem_edges:** The two files provide
the mean strength of each susbsystem edge, separately for alpha and theta.
This is provided in a four column tabular format with the last column
providing the mean edge strength (average iPLV value) for the given
subject, condition, and subsysyem edge. ### **Figure_4** **6_8-05:**
Inside the subfolder ‘c_limit(0.06)’, each of the four files contain
group-level connectomes in the theta-band. The AM 0, 1, 2, and 99 refer to
the Shape-specific, Color-specific, Location-specific, and Shared
networks, respectively. Each connectome is a 200 x 200 adjacency matrix.
**11-25_13-06**: Same as above but for the alpha-band ### **Figure_5**
**Correlation:** Each file contains the K values for each frequency for
the correlation between phase synchronization and behavioural performance.
The naming of the files as well as the K values within each file follow
the convention used in ‘Figure 2/Contrast_betn_features’.
**Graph_strength:** Each file provides the graph strength for each feature
(Shape, Color, Spatial) and frequency band (alpha, theta). Within each
file, each row represents a subject and the third column, iPLV, contains
the mean iPLV for the significant network. ### **Figure_6** **Amplitude:**
For each subject, and separately for the retention and baseline window,
classification accuracy in decoding the memorized feature of each trial.
Files are separate for each subject (S0**), load (2 or 4), and whether
baseline (BL) window was used or not. In each file, the decoding accuracy
for Early and Late is provided in separate columns, with decoding accuracy
being the proportion of trials accurately classified. **Synchrony:** Same
as above but for phase synchronization. ### **Figure_7** **PAC_python:**
Each file contains the K values for significant phase-amplitude coupling
of the inter-areal networks in each of the visual features. K values for
each n:m ratio (e.g. 1-2, 1-3) are provided in separate files. The naming
of the files as well as the K values within each file follow the
convention used in ‘Figure 2/Wilcoxon/jackknife’. ## **Subject folders
(S0****) Each of the twenty subject folders (from S001, S002, … S020)
follow the same folder and file structure. There are three folders inside
each S0** folder. **3x2_Feature_x_Obj:** contains the connectomes of phase
synchronization for each feature (Shape, Color, Location) and separated by
load (load 2 or load 4). **3x2_Load_avg:** contains the connectome of
phase synchronization for each feature (Shape, Color, Location) averaged
across the two loads. **3x2_Load_avg_PAC:** contains the connectome of
phase-amplitude coupling for each feature (Shape, Color, Location). In
each of the three folders, the connectome data at the subject level is
provided in .tdms files. The filename structure for each TDMS file is as
follows: EdgeDAmplitude-Phase *ratio* Original No-Surrogates cPLV Lag_1-0
Lag=1.000 Low = *low_frequency* *condition* *folder_name*
*high_frequency*Hz.tdms *ratio*: for 3x2_Feature_x_Obj and 3x2_Load_avg,
the *ratio* is always 1-1. In 3x2_Load_avg_PAC, since the connectomes
reflect n:m synchronization, the *ratio* would be different based on the
values of *low_frequency* and *high_frequency* *low_frequency* and
*high_frequency*: for 3x2_Feature_x_Obj and 3x2_Load_avg, *low_frequency*
and *high_frequency* are the same while they are different for
3x2_Load_avg_PAC. This indicates the frequency in which the source
reconstructed data was narrow-band filtered to before the synchronization
estimates were obtained. *condition:* this indicates the visual feature in
which the synchronization was estimated from. The condition is either
Shape_HIT, Color_HIT, or Spatial_HIT, each corresponding to a visual
feature. For 3x2_Feature_x_Obj, the load is also provided, e.g.
Shape2_HIT. *folder_name:* this is the folder in which the file is located
and is either 3x2_Feature_x_Obj, 3x2_Load_avg, or 3x2_Load_avg_PAC. Within
each TDMS file the subject-level connectome is provided for the specific
condition and frequency band. The connectome for the baseline, early
retention, and late retention are provided in the groups 0, 1, and 2,
respectively. Each connectome is a 400 x 400 adjacency matrix with each
edge containing the iPLV value. ## Human subjects data We have received
explicit consent from the participants that de-identified data could be
published in the public domain. To de-identify the data, all personal
information that may be used to identify a participant has been removed.
Each subject is only labelled via a numbered code (e.g. S001) and subject
data only contains connectome data (400 x 400 matrices of complex valued
numbers). Any physiological data that could be used to identify the
subjects (e.g. raw EEG/MEG traces, MRIs etc.) are not included in the
dataset.