Knights et al. (2024). Neural Evidence of Functional Compensation for Fluid Intelligence in Healthy Ageing.

This repository contains code accompanying the multivariate Bayesian machine-learning article published in eLife:
Knights, E., Henson, R. N., Morcom, A. M., Mitchell, D. J., & Tsvetanov, K. A. (2024). Neural Evidence of Functional Compensation for Fluid Intelligence in Healthy Ageing. eLife. https://doi.org/10.7554/eLife.93327.1

Modelling Quickstart

Clone & download processed fMRI/behavioural data with shell from root:

git clone https://github.com/ethanknights/Knightsetal_fMRI-Cattell-Compensation

cd Knightsetal_fMRI-Cattell-Compensation

./setup.sh

Use R to perform univariate/multivariate modelling analysis (see README sections 2/4 in 'Analysis Pipline') e.g.:

setwd('R')

source('load_data.R')

source('plot_ROI.R') #For interaction: activation ~ age * bhv

Prerequisites

Software

• FSL v4.0+
• Matlab (tested with r2019a)
  • SPM12
    https://www.fil.ion.ucl.ac.uk/spm/software/spm12/
  • Commonality Analysis tools
    https://github.com/kamentsvetanov/CommonalityAnalysis
  • BrainNet toolbox
    https://www.nitrc.org/projects/bnv/
• R
  • MASS
  • BayesFactor
  • ggeffects
  • Popular manipulation/visualisation packages (e.g., ggplot2, dplyr, tidyverse)
• Python
  • gl package (to plot t-statistic maps in mricroGL)

Datasets

Raw data for Machine-Learning

• For performing Multivariate Bayes and the preceding whole-brain univariate analysis (e.g. generating second-level models, defining ROIs, extracting timeseries), download the Cattell first-level model images derived from automatic analysis (see Taylor et al. 2017 NeuroImage) by submitting a Cam-CAN data-request:
  https://camcan-archive.mrc-cbu.cam.ac.uk/dataaccess
  
  

Summary Data for Modelling

• For hypothesis testing using R, download data from the Open Science Framework (OSF).

Linux/Mac

./setup.sh  #Download, extract & move the summary datasets from the OSF, to their required destinations

Windows

• Manaully download/extract the summary_data.zip from https://osf.io/v7kmh/ and move each .csv to the following corresponding analysis directories:

Univariate

mkdir R/csv
mv univariate.csv R/csv/

Univariate (Vascular control)

mkdir R_vascular/csv
mv univariate_vascular_RSFA.csv R_vascular/csv/

MVB

# Cuneal Cortex
mkdir MVB/R/dropMVBSubjects-1/70voxel_model-sparse/csv
mv MVB_cuneal.csv  MVB/R/dropMVBSubjects-1/70voxel_model-sparse/csv/

# Frontal cortex
mkdir MVB/R/dropMVBSubjects-1/control-frontalROI_70voxel_model-sparse/csv
mv MVB_frontal.csv MVB/R/dropMVBSubjects-1/control-frontalROI_70voxel_model-sparse/csv

MVB (Multivariate mapping)

# Cuneal Cortex
mv MVB_multvariateMapping-Cuneal/extradata_ShuffledGroupFVals.csv MVB/R/dropMVBSubjects-1/70voxel_model-sparse/csv/

# Frontal cortex
mv MVB_multvariateMapping-Frontal/extradata_ShuffledGroupFVals.csv MVB/R/dropMVBSubjects-1/control-frontalROI_70voxel_model-sparse/csv

Analysis Pipeline

1) Whole-brain Univariate Analysis

Use matlab to run fMRI preprocessing and generate the group univariate contrasts (i.e., second level model generation, TFCE correction, ROI definition and timeseries extraction):

pipeline.m

/* Alternatively, download the summary datasets described earlier in the Prerequisties section.

Use Python to plot TFCE-corrected t-statistic maps with MRIcroGL.

./python/mosaic_MRICRON_HardBIGEasy.py

gl.overlayload('./nii_HardBIGEasy/Intercept_tfce197.nii')
gl.minmax(2, 0, 10000)
gl.colorname (1,"1red")

./python/mosaic_MRICRON_Covariates.py:

gl.overlayload('/nii_Covariates/tval_Age_tfce197.nii')
gl.minmax(1, 0, 250)
gl.colorname (1,"2green")

gl.overlayload('nii_Covariates/tval_PC6_tfce197.nii')
gl.minmax(2, 0, 250)
gl.colorname (2,"1red")

2) Univariate Modelling

In R, generate the regression models & plots:

setwd('R')
source('loadData.R')

#For interaction: Bhv ~ age * gender
source('plot_bhv.R')

#For interaction: Activation ~ age * bhv
source('plot_ROI.R')

#For the control vascular RSFA analysis:
setwd('R_vascular')
source('load_data_RSFA.R')
source('plot_ROI_RSFA.R')

3) Multivariate Bayes

Use matlab to run the MVB machine-learning pipeline:

cd MVB

setupDir.m
wrapper.m
doPostProcessing.m

/* Alternatively, download the summary datasets described earlier in the Prerequisties section.

4) Multivariate Modelling

In R, generate the logistic regression models & plots for the ROI datasets independently:

#CHOOSE ONE ROI: 
setwd('./MVB/R/dropMVBSubjects-1/70voxel_model-spare') #Cuneal Cortex
setwd('./MVB/R/dropMVBSubjects-1/control_frontalROI_70voxel_model-spare') #Frontal Cortex

#Load data into dataframe 'df'
source('run_loadData.R')

For testing the multvariate mapping, use raincloud package:

source('run_MVB_shuffledMapping.R') 

For logistic regression of Boost likelihood, use MASS/polr packages:

source('run_fMRI_MVB.R')

How to Acknowledge

Please cite:
Knights, E., Henson, R. N., Morcom, A. M., Mitchell, D. J., & Tsvetanov, K. A. (2024). Neural Evidence of Functional Compensation for Fluid Intelligence in Healthy Ageing. eLife. https://doi.org/10.7554/eLife.93327.1