This repository contains a set of scripts used 1) to analyze the relationship between amyloid-beta accumulation, functional connectivity and memory and 2) to analyze the effects of education on functional connectivity and memory in the OASIS-3 cross-sectional data.
analysis_ICA25d.m includes scripts that import the data, exclude several participants identified by QC; and run linear models (fitlm) to test for the effects of education on pairwise functional connectivity. The functional connectivity variables undergo exclusion of outliers (>3SD from the mean) first to ensure data quality is adequate.
Next , we use 1,000 permutations to test for significance of the effect of education. Each permutation involves reshuffling the outcome variable (of the 210 functional connectivities). From each permuted model we get a "by chance" F-value which allows us to generate a permutation distribution.
We provide code for visualizing the connectivity matrices and circular graphs.
pls_ICA25d.m set of scripts includes code that imports and cleans up data similar to ICA25d.m. USing the 210 connectivities as predictors (X), and age and delayed memory recall as outcomes (Y), partial least squares PLS regression is ran. Permutation testing n=5,000 is used to test whether the PLS explains a significant amount of variance in the outcome (Y) and bootstrapping is used to test which connectivities show a robust contribution (Z score > 3) to the latent PLS variable.
3. Functional connectivities mediate the relationship between mean cortical amyloid-beta and memory recall
A PLS structural equation model script is used here (pls_sem_O3.m). PLS-SEM toolbox outputs are provided in the folder called report_FINAL_BL for the baseline data and report_extended_sample for the extension analysis. https://www.mathworks.com/matlabcentral/fileexchange/54147-pls-sem-toolbox
Scripts used to run the analyses in points 2. and 3. (i.e. the PLS and PLS-SEM) in the extension sample are called pls_ICA25d_m.m and pls_sem_O3.m, respectively.
Please note that we do not include the PLS-SEM analyses in our manuscript but share the code for the PLS-SEM nevertheless in case it might be useful.