This repository contains two datasets from perceptual learning studies using the Glass Pattern task: a multi-session training study and a tDCS intervention study. MRI data were collected using resting-state fMRI and quantitative multi-parameter mapping techniques. Genetic analyses were performed using gene expression data from the Allen Human Brain Atlas (AHBA). The repository includes the toolboxes used to generate the data, along with the source data corresponding to each figure in the manuscript.

repo27032025.csv Figure1 sheet: Corresponds to Figure 1 in the manuscript, detailing the toolboxes employed in the methods. A. Gene expression data were downloaded and analyzed using the Abagen toolbox,https://abagen.readthedocs.io/en/stable/. B. Functional and microstructural gradient preprocessing was performed using the micapipe toolbox, https://micapipe.readthedocs.io/en/latest/index.html. D. E.Gene expression enrichment analysis was conducted using the DAVID website (https://davidbioinformatics.nih.gov/). For specific toolbox configurations, please refer to the Methods section of the manuscript.

Figure2 sheet: Contains behavioral performance data from the multi-session training experiment. Each row represents a participant, and each column corresponds to a different training session (Day 1, 5, 6, 7, 8, and 9). Scan_1: Baseline MRI, Scan_2: Pre-training MRI, Scan_3: Post-training MRI.

Figure3 sheet:Contains MRI data from the multi-session training experiment. The top table presents functional connectivity dispersion within networks (EV – visual, FPN – frontoparietal) and between networks (EV-FPN). The bottom table presents microstructural dispersion data. Behavioral performance is reported as the percentage improvement in accuracy following training.

Figure4 sheet: Contains data from the tDCS experiment. The top table shows behavioral accuracy for each group (Anodal, Sham) before and after the tDCS intervention. The bottom table shows functional connectivity dispersion (within-network EV and FPN; between-network EV-FPN) for the Anodal group, normalized against the Sham group, before and after the intervention.

This data supports the publication 'Cellular and molecular signatures of in vivo imaging measures of GABAergic neurotransmission in the human brain' in Communications Biology: https://rdcu.be/cLDWQ.

The Binding_PLSresult file contains radiotracer binding values of the above mentioned [11C]Ro15-4513 parametric map, resampled into the Desikan-Killiany atlas space using the fslmeants function from FSL. Additionally, it contains the weights of partial least square regression analysis (PLS) performed on the radiotracer binding data (resampled [11C]Ro15-4513 and [11C]flumazenil radiotracer binding parametric maps) and the above mentioned gene expression dataset (gene-wise and cluster-wise). PLS was performed with an existing script (https://github.com/SarahMorgan/Morphometric_Similarity_SZ) run in Matlab R2017a. Gene expression data from left hemisphere only were included due to the low number of participants included in the right hemisphere dataset. PLS comprised a linear analysis of covariance between the gene expression and radiotracer binding data with the use of a principal component analysis for dimension reduction of the gene expression dataset with 1,000 permutations accounting for spatial autocorrelations as described in previous publications. Bootstrapping was performed to calculate the Z-scores and hence the rank of each gene (or cluster) contribution to the result. The Ro15_template is an average parametric map of [11C]Ro15-4513 binding in 10 healthy volunteers (four females, mean age +/- SD 25.40 +/- 3.20, range 22-30). The study was approved by the London/Surrey Research Ethics Committee. All subjects provided written informed consent prior to participation. Data were acquired on a SignaTM PET-MR General Electric (3T) scanner using the MP26 software (01 and 02) at Invicro, A Konica Minolta Company, Imperial College London, UK. PET acquisition was performed in 3D list mode for 70 minutes. A ZTE sequence was used for attenuation correction (voxel size: 2.4x2.4x2.4mm3, field of view=26.4, 116 slices, TR=400ms, TE=0.016ms, flip angle=0.8o) and PET image co-registration was performed with a T1-weighted IR-FSPGR sequence (voxel size: 1x1x1mm3, field of view=25.6, 200 slices, TR=6.992ms, TE=2.996ms, TI=400ms, flip angle=11o). Individual subject images were generated with MIAKAT v3413 in Matlab R2017a through a simplified reference tissue model using the pons as the reference region and solved with basis function method. The individual parametric maps were averaged using SPM imCalc function.

The WGCNA_result R data file contains the result of Weighted Gene Co-Expression Network Analysis on 15,633 downloaded from the Allen Human Brain Atlas. The data was downloaded with the abagen toolbox in JupyterLab Notebook through anaconda3 in Python 3.8.5. The data was thresholded with an intensity-based filter removing probes with intensity less than background in half or more of the samples, and mapped onto 83 brain regions based on the Desikan-Killiany Atlas. WGCNA was performed in R 4.0.3 with the ‘signed’ method and soft threshold power=14. Individual modules were isolated with the classic ‘tree’ dendrogram branch cut.