Table of Contents

1. Introduction
2. Prerequisites
3. Setup
4. Input Data Format
5. Usage in Python, MATLAB, and R
6. Function Parameters
7. Supported Disorders
8. Global Mean Correction (gmean)
9. Supported Atlas Parcellation Schemes
10. Output PCS Scores
11. Citing the PCS toolbox
12. License

Introduction

The PCS toolbox calculates polyconnectomic score (PCS) to quantify the presence of disease-related brain connectivity signatures in individual connectomes. PCS integrates existing knowledge of disorder-related brain circuitry, aggregating these connectivity signatures or connectome summary statistics (CSS) across the entire brain into a single, interpretable metric.

This repository includes toolbox scripts and CSS of resting-state functional connectivity (FC) signatures for various neuropsychiatric disorders. These CSS represent the strength and direction of associations between brain connections and specific disorders across connectomes. The CSS were generated through the following process: - Effect size estimation per dataset: Cohen’s d was computed for each FC, corrected for age, sex, in-scanner motion, and site, generating a connectivity matrix for each dataset within a disorder. This matrix indicates connections with hypo-connectivity in patients (negative d) or hyper-connectivity in patients compared to controls (positive d). - Meta-Analysis: these connectivity-wise effects were aggregated across datasets for each disorder using a meta-analytic approach with a random-effects model.

We provide the meta-analytic CSS for 11 neuropsychiatric and neurological disorders, encompassing 10,667 individuals (5,325 patients, 5,342 unique controls) across 22 datasets, allowing researchers to compute PCS on their own data.

Prerequisites

• Python 3.7+, MATLAB, or R #### Required packages:
• Python: NumPy >= v1.21.0
• R: No additional packages required
• MATLAB: No additional toolboxes required

Setup

Clone this repository:

git clone https://github.com/dutchconnectomelab/pcs-toolbox.git

For R Users

1. Download Required Files: Similarly, download the R script calculate_PCS.R from this repository.

2. Source Script: Use the following command within R to source the calculate_PCS.R script for use: ```r source("/path/to/calculate_PCS.R")

For MATLAB Users

1. Download Required Files: Instead of cloning the repository, download the MATLAB script calculate_PCS.m directly.

2. Add Path: Add the directory containing calculate_PCS.m to MATLAB's path using addpath('/path/to/folder/containing/calculate_PCS').

Installation Dependencies

In Python, ensure you have the following packages installed:

• NumPy

You can install these packages in python using pip:

bash pip install numpy

Input Data Format:

The cnn parameter is a connectivity matrix representing functional brain connections, typically a correlation matrix. The function calculate_PCS automatically detects the dimensions of the input cnn matrix to determine the number of regions of interest (ROI) and subjects:

• 2D Matrix (Nregions × Nregions): If cnn is a 2D matrix, PCS is calculated assuming it represents the connectivity between ROIs for a single subject.

• 3D Matrix (Nregions × Nregions × N_subjects): If cnn is a 3D matrix, PCS is calculated for each subject represented in the third dimension; computed individually for each subject's connectivity data.

This allows for flexibility in handling both single-subject and multi-subject connectivity data. It is crucial to ensure that the atlas used to generate the input connectivity matrix matches the atlas specified in the function parameters for accurate PCS calculation.

Ensure consistency in dimensions and region order

To successfully calculate PCS, the connectivity matrix of a subject must match the dimensions of the CSS for the specified atlas. If the dimensions do not align, PCS calculations will not function correctly. Additionally, it is crucial that the order of regions in the connectivity matrix corresponds exactly to the order in the CSS. You can verify the order of regions by checking the index and columns of the CSS files provided. Please make sure to verify these details before running your analysis.

Usage in R, MATLAB, and Python

R

```r

Source the calculate_PCS.R script from your cloned repository:

source("path/to/your-repository/calculate_PCS.R")

Load RDS file with connectivity data

cnn <- readRDS("path/to/your-connectivity-matrix.rds")

Calculate the PCS for schizophrenia using Desikian-Killiany atlas and global mean corrected

PCSsubject <- calculatePCS(cnn, disorder='schizophrenia', gmean=TRUE, atlas='aparc') ```

MATLAB

```matlab % Add the directory containing calculatePCS.m to MATLAB's path: addpath('path/to/your-repository') % Load .mat or .csv connectivity data cnn = load('path/to/your-repository/connectivitydata.mat');

cnn = readmatrix('or/path/to/connectivity_data/in/csv/format.csv');

Calculate the PCS for schizophrenia using Desikian-Killiany atlas and global mean corrected

PCSsubject = calculatePCS(cnn, 'disorder', 'schizophrenia', 'gmean', true, 'atlas', 'aparc'); ```

Python

```python

Install the required Python packages:

pip install numpy

Ensure the calculate_PCS.py script is available in your working directory or in your Python path:

sys.path.append('path/to/your-repository')

import numpy as np from calculate_PCS import *

Load .npy connectivity data

cnn = np.load('path/to/your-repository/connectivity_data.npy')

Calculate the PCS for schizophrenia using Desikian-Killiany atlas and global mean corrected

PCSsubject = calculatePCS(cnn, disorder='schizophrenia', gmean=True, atlas='aparc') ```

Function Parameters

• cnn: User connectivity data (2D or 3D array)
• disorder: Name of the disorder (string)
• gmean: Boolean, whether to use functional global mean normalization
• atlas: Name of the brain atlas used in both cnn and CSS creation

The PCS toolbox supports the following neuropsychiatric and neurodegenerative disorders:

• Alzheimer's disease: "alzheimer"
• Anxiety-related disorders: "anxiety"
• Attention-deficit/hyperactivity disorder: "adhd"
• Autism spectrum disorder: "autism"
• Bipolar disorder: "bipolar"
• Frontotemporal dementia: "ftd"
• Major depressive disorder: "depression"
• Obsessive-compulsive disorder: "ocd"
• Parkinson's disease: "parkinson"
• Schizoaffective disorder: "schizoaffective"
• Schizophrenia: "schizophrenia"

When using the calculate_PCS function, specify the disorder parameter using its key (e.g., 'adhd', 'ftd', 'schizophrenia', etc.).

The demographics of the samples used to compute the CSS per disorder are as follows:

| Disorder | Patients, n | Controls, n | Age (Mean ± SD) | Male/Female, n | Datasets, n | |------------------------------------------|---------------|---------------|-------------------|------------------|---------------| | Alzheimer's disease | 270 | 866 | 72.0 ± 8.9 | 510/626 | 4 | | Anxiety-related disorders | 366 | 840 | 27.1 ± 21.3 | 505/701 | 3 | | Attention-deficit/hyperactivity disorder | 1211 | 1417 | 19.1 ± 16.7 | 1518/1110 | 4 | | Autism spectrum disorder | 1151 | 2086 | 21.6 ± 14.3 | 2530/707 | 4 | | Bipolar disorder | 320 | 1790 | 36.3 ± 13.8 | 975/1135 | 5 | | Frontotemporal dementia | 156 | 143 | 64.0 ± 8.5 | 144/155 | 2 | | Major depressive disorder | 884 | 1605 | 33.9 ± 15.8 | 1122/1367 | 4 | | Obsessive-compulsive disorder | 98 | 261 | 12.7 ± 4.5 | 170/189 | 3 | | Parkinson’s disease | 102 | 125 | 66.4 ± 8.5 | 110/117 | 4 | | Schizoaffective disorder | 235 | 812 | 37.2 ± 12.8 | 403/644 | 3 | | Schizophrenia | 532 | 1882 | 36.5 ± 13.6 | 1254/1160 | 6 |

Global Mean Correction (gmean)

Global mean signal correction is a technique used to reduce the impact of non-neural physiological noise in fMRI data, although its use is debated in the neuroimaging community as it can also remove some neural signals of interest. The PCS toolbox provides options for both corrected and uncorrected CSS, allowing researchers to choose based on their specific research questions and methodological preferences.

• When gmean=True (Python) or gmean is included (MATLAB/R), the toolbox uses functional CSS that have been corrected for global mean signal. This correction involved regressing out the framewise mean signal intensity of all brain voxels.

• When gmean=False (Python) or the parameter is omitted (MATLAB/R), the toolbox uses functional CSS without global mean signal correction.

We advise using CSS with gmean if the provided cnn also uses this preprocessing method.

The PCS toolbox supports the following atlas parcellation schemes:

• aparc (Desikan-Killiany) (68 regions)
• BB50human (76 regions)
• Brodmann (78 regions)
• brainnetome (210 regions)
• Campbell (34 regions)
• EconomoCT (30 regions)
• economo (86 regions)
• Flechsig (92 regions)
• hcp-mmp-b (Glasser) (360 regions)
• Kleist (98 regions)
• lausanne120 (114 regions)
• lausanne250 (219 regions)
• lausanne500 (448 regions)
• nspn500 (308 regions)
• schaefer100-yeo7 (100 regions)
• schaefer300-yeo7 (300 regions)
• schaefer400-yeo7 (400 regions)
• schaefer500-yeo7 (500 regions)
• schaefer600-yeo7 (600 regions)
• schaefer700-yeo7 (700 regions)
• schaefer800-yeo7 (800 regions)
• schaefer900-yeo7 (900 regions)
• schaefer1000-yeo7 (1000 regions)
• Smith (88 regions)
• yeo17dil (116 regions)

Output PCS Scores:

The calculate_PCS function returns an array of PCS scores, one for each subject in the input data. Higher PCS indicate a stronger presence of connectivity patterns associated with the specified disorder. Importantly, scores are relative measures and should be interpreted in comparison to a control group or normative data.

Datasets

Datasets used in the creation of CSS can be accessed via their respective websites:

Openly Accessible Datasets - Autism Brain Imaging Data Exchange (ABIDE-I & ABIDE-II): fcon_1000.projects.nitrc.org/indi/abide - Advancing Research and Treatment for Frontotemporal Lobar Degeneration - Longitudinal Evaluation of Familial Frontotemporal Dementia Subjects (ARTFL-LEFFTDS): memory.ucsf.edu/research/studies/nifd - ADHD-200 Sample: nitrc.org/frs/?group_id=383 - Alzheimer's Disease Neuroimaging Initiative (ADNI-2/GO & ADNI-3): ida.loni.usc.edu - Boston Adolescent Neuroimaging of Depression and Anxiety (BANDA): (NDA; NDAR ID: 3037) - Bipolar & Schizophrenia Consortium for Parsing Intermediate Phenotypes (B-SNIP1): (NDA; NDAR ID: 2274) - Bipolar & Schizophrenia Consortium for Parsing Intermediate Phenotypes 2 (B-SNIP2): (NDA; NDAR ID: 2165) - Latin American Brain Health Institute (BrainLat): synapse.org - Centre for Biomedical Research Excellence (COBRE): schizconnect.org - Consortium for Neuropsychiatric Phenomics (CNP): OpenNeuro openneuro.org - Healthy Brain Network (HBN): fcon1000.projects.nitrc.org/indi/cmihealthybrainnetwork - Enhanced Nathan Kline Institute (NKI-Enhanced): fcon_1000.projects.nitrc.org/indi/enhanced/ - Open Access Series of Imaging Studies (OASIS-3): oasis-brains.org - Brain Function and Genetics in Pediatric Obsessive-Compulsive Behaviors (OCD-Pediatric): (NDA; NDAR ID: 2955) - Task Control Circuit Targets for Obsessive Compulsive Behaviors in Children (OCD-Task): (NDA; NDAR ID: 3044) - Parkinson’s Disease Yonemaya: openfmri.org - Parkinson’s Disease Tao Wu and NEUROCON: fcon_1000.projects.nitrc.org/indi/retro/parkinsons.html - Japanese Strategic Research Program for the Promotion of Brain Science (SRPBS): bicr-resource.atr.jp/srpbs1600

Restricted Access Datasets - Marburg-Münster Affective Disorders Cohort Study (FOR2107): for2107.de

Citing the PCS toolbox

When using PCS toolbox, please cite the following paper: Libedinsky, I., Helwegen, K., Boonstra, J., Guerrero Simón, L., Gruber, M., Repple, J., et al. (2024). Polyconnectomic scoring of functional connectivity patterns across eight neuropsychiatric and three neurodegenerative disorders. Biological Psychiatry. https://doi.org/10.1016/j.biopsych.2024.10.007

When computing PCS for a particular disorder using the precomputed CSS, please cite the corresponding papers listed in Citations. These sources were used for computing the CSS.

License

pcs-toolbox © 2024 by dutchconnectomelab is licensed under CC BY-NC 4.0. To view a copy of this license, visit https://creativecommons.org/licenses/by-nc/4.0/