Knights et al. (2021). Does Hemispheric Asymmetry Reduction in Older Adults (HAROLD) in motor cortex reflect compensation? Journal of Neuroscience, (45), 9361-9373.

About

The repository contains code (R, Matlab, Python & Shell) to replicate the machine learning big-data report published in the Journal of Neuroscience (https://www.jneurosci.org/content/41/45/9361)

Prerequisities

• R
  • MASS
  • ggplot2
  • brms
  • bayesfactor
• Matlab
  • SPM
• Python
  • sklearn
  • numpy
• Datasets
  • For ROIs (definition/timeseries extraction) & multivariate procedures (MVB, MVPA):
    • MVB Sensorimotor Task fMRI & T1-weighted image release004 datasets (request from www.cam-can.org).
  • For R modelling:
    • Download preprocessed data .csv's (www.osf.io/seuz5)

1. f/MRI Preprocessing

An Automatic Analysis (github.com/automaticanalysis) pipeline preprocessed the Stage 3 Cam-CAN FreeSelection dataset using automaticanalysis. In matlab:

cd freeSelection/aa_cc280_fingerTapping
run_aa

Standard SPM models can then be generated using matlab:

cd SMT/pp
wrapper

Ensure the necessary switches are set to false. e.g. If running from scratch:

done_setupDirs = true; /*switch to false if rerunning

2. ROI definition

During the createROI.m routine, use SPM_results_ui to load data/groupGLM/SPM.mat. Next, select the peaks of brain activation in right and left motor cortex and create the binary masked Regions of Interest (ROIs):

createROI /*outputs: PreCG_R_70.nii,PreCG_L_70.nii etc.

3. Machine learning: Multivoxel Pattern Analysis (MVPA) & Multivariate Bayesian Decoding (MVB)

The convenience wrapper script (SMT/pp/wrapper.m) concludes by extracting beta estimates for single trials to single subject *.mat files.
For machine learning, follow the MVB functions/README via the SMT/MVB/ routine.

For MVPA (using SVM classification), follow the Matlab routine (freeSelection/pp/classify_run.m) which can be validated using an additional sk-learn Python routine (freeSelection/pp/classify.py)

4. R: Brain-Behaviour Modelling

The MVB sub-directories contains matlab code to create the data-tables for R analysis (SMT/MVB/doPostProcessing.m) though this post-processing script will fail to collect behavioural phenotype data (e.g. Reaction Times) unless your workstation is connected to the MRC CBU environment.

For convenience, the processed data-tables can be downloaded from the Open Science Framework (www.osf.io/seuz5).

Move these data.csv's to the appropriate csv directory (e.g. for the analysis reported in Figure 4, place the data.csv in SMT/MVB/R/70voxel_model-sparse/csv/data.csv).

In R, load the appropriate data.csv with SMT/MVB/R\<analysistype\>/<analysistype\>/run_001_loadData.R.
From there, any of the classical or Bayesian analyses can be performed using the dataframe: df.

For example, to plot the hyperactivation of motor cortex in older adults:

setwd('HAROLD-MVB/tree/main/SMT/MVB/R')
source('run_fMRI_univariate.R')
#This boils down to writing a point ggplot based on the following robust linear regression model:
rlm_model <- rlm(univariateMean_R ~ age0z2, 
                 data = df, psi = psi.huber, k = 1.345)

How to Acknowledge

Please cite: Knights, E., Morcom, A., & Henson, R. N. (2021). Does Hemispheric Asymmetry Reduction in Older Adults (HAROLD) in motor cortex reflect compensation? Journal of Neuroscience, (45), 9361-9373.