Functional Connectivity Changes in Aging and Alzheimer's Disease

This repository provides code for reproducing analyses and figures from the manuscript:

"Age and Alzheimer’s disease affect functional connectivity along separate axes of functional brain organization"

Preprint

Due to data-sharing restrictions, the repository uses synthetic data by default.

ℹ️ NOTE

The source code in the repo is for the final version of the manuscript, the code in the docker container produces the version of the manuscript that is published as a preprint. You won't be able to run all of the source code in the current repo with the synthetic data. As soon as I have time I will synthesize new data and update everything so that you can run everything from start to finish for the final version.

---

Getting Started

You can either clone the repo as it is and try to set it up using in R using renv::restore() and then source src/main.R, but the easiest way to reproduce the analyses and manuscript is likely to use a docker container with the dependencies.

Docker Image

First install docker https://www.docker.com/get-started/

The Docker image is available on Docker Hub:

docker pull jorittmo/fc-changes-image:latest

Running the Analysis with Docker

To reproduce the paper using the synthetic data provided clone the repo to a directory of your choice and then run:

docker run --rm \
  -v /fc_changes_follow_gradients/data/adni_src_data_synthetic:/fc_changes/data/adni_src_data_synthetic \
  -v /fc_changes_follow_gradients/data/bf_src_data_synthetic:/fc_changes/data/bf_src_data_synthetic \
  -v /fc_changes_follow_gradients/docker_out:/fc_changes/paper/output \
  -e FROM_START=FALSE \
  jorittmo/fc-changes-image \
  Rscript src/main.R

This will run all analyses (without pre-calculating measures etc) and spit out the manuscript with figures usin synthetic data in docker_out.

Ensure the host mount paths are the absolute path to the repo (i.e. ~/fc_changes_follow_gradients, if you put it in you home folder).

• -v mounts local directories into the container for data input and output.
• -e FROM_START sets if you want to do all pre-computation to get derivatives, gradients and such. Set to TRUE to recompute everything from scratch and to FALSE to only run analyses and models and generate figures.

If you want to try the image without running the full analysis pipeline which takes ~10-30 minutes depending on setup, and if you run it from start you can just render the paper by doing:

docker run --rm \
  -v /fc_changes_follow_gradients/docker_out:/fc_changes/paper/output \
  -e FROM_START=FALSE \
  jorittmo/fc-changes-image \
  Rscript src/render_paper.R

The rendered paper (with synthetic data) should then be reachable in the folder docker_out.

---

Configuration and Arguments

The script main.R recognizes the following environmental arguments:

Argument	Description	Recommended Setting (Synthetic Data)
FROM_START	Recompute everything (connectome calculation, FC derivatives, gradients) from scratch	FALSE (data precomputed)
CREATE_BRAIN_PERMUTATIONS	Generate brain permutations for spin tests (takes quite long to run and is not necessary)	FALSE (already included)
EXTRACT_TIMESERIES	Extract parcel time series from raw images (should not be used for now)	FALSE (timeseries pre-extracted)
REAL_DATA	Toggle between real and synthetic data (Should not be used for now)	FALSE (synthetic data)

To override the defaults, set them as environmental variables when calling Docker (e.g., -e FROM_START=TRUE).

---

Workflow Summary

The code in main.R runs the following major steps:

1. Initial setup

   • Set flags (FROM_START, CREATE_BRAIN_PERMUTATIONS, EXTRACT_TIMESERIES, REAL_DATA), create folders.

2. Atlas Loading

   • Load/parcellation (Schaefer, Yeo), build ROI-to-network lookup, gradient color palettes.

3. (Optional) Permutations

   • If requested, generate 1,000 spin/test parcellation rotations and save them.

4. (Optional) Timeseries Extraction

   • If requested, use Nilearn to read raw NIfTIs, extract 1,000-parcel time series, save as .rds.

5. Load & Clean BioFINDER Data

   • Read CSV → filter for motion & diagnosis/pathology → compute continuous pathology trajectories via SCORPIUS → finalize biofinder_df.

6. (FROM_START) Build BioFINDER Connectomes & Derivatives

   • Read saved timeseries → compute 1000×1000 correlation matrices → save → read them back into a big 3D array → compute node-strength + two "affinity” measures via fc_strength/get_affinity → save those results.

7. (FROM_START) Compute BioFINDER Gradients & Parameter Sweep

   • Load “Margulies” reference gradients (CSV or via Nilearn), build “healthy-young” average connectome → plot supplementary gradient-comparison figure and save → call get_gradients once at (threshold=0, method=pca; the gradients used) → then loop over several{threshold, method, affinity} combos to build a large dataframes with gradients derived over different values.

8. Load & Clean ADNI Data

   • Read CSV → rename columns → compute continuous pathology_ad via SCORPIUS → load FD CSVs (not computed in this workflow) → (if FROM_START) extract/scrub timeseries → build ADNI connectomes → filter by motion → save cleaned adni_df.

9. (FROM_START) Compute ADNI Connectivity Derivatives

   • Read 3D array of ADNI connectomes → compute node-strength + two affinity variants → save.

10. (FROM_START) Compute ADNI Gradients & Merge

    • Build “young, healthy” average ADNI connectome → call get_gradients once (PCA) → bind with BioFINDER gradients → loop over parameter grid to build gradient_data_adni → merge with BioFINDER → save a

11. (Optional) Methods Figure

    • Run code to generate pngs used in the methods plot, this is now commented out in main.R, just uncomment our source methods_figure.R to generate the figures used in it. This plot is not reproducible as it was put together in inkscape. All other figures are generated.

12. Main Figures

    • Figure 1: figure_one(...) cross-sectional biofinder + ADNI (cognition is also generated but we take only the first figure panels), save as PNG
    • Figure 2: first calculate non-linear trajectories: gam_pred_nodes(...) → plot_gams_v1(...), save as PNG
    • Figure 3: Build "longitudinal" df → longitudinal_and_window_analysis(...), save main & supp PNG
    • Figure 4: figure_one(...) cognition panels only, save as PNG

13. Supplementary Figures

    • Many calls to plot_gradient_relationships(...), figure_one(...), overlaid_main_results(...) with different and/or measures (diagnosis, gradient 2 only, within/between-network affinity, nodal strength, no-threshold affinity, clinical interactions, healthy group interactions, pathology densities etc.). Each saved to a distinct file under paper/figures/….

14. Table 1 (Cross-Sectional)

    • Build descriptive finalfit tables for BioFINDER & ADNI, row-bind, save as CS_tbl1.rds.

15. Longitudinal Table 1 (BioFINDER)

    • Compute baseline/follow-up summary for each BioFINDER subject → summary_factorlist(...), save as LT_tbl1.rds.

16. (FROM_START) Gradient Sensitivity Analyses (Supplementary Table)

    • If FROM_START, call write_gradient_supp_table(params), save as supp_table.rds; otherwise just load.

17. Quarto Rendering

    • Source the script render_paper.R which calls quarto to render the manuscript and then moves the rendered paper into paper/output

---

Troubleshooting

If you encounter any issues, please open an issue on GitHub, and we'll assist you as quickly as possible.

The non-container version of this code has only been tested locally on ubuntu 22.04. If you have dependency troubles I recommend the containerized version.

The code is not as well documented as I would have wanted at this time, so please raise an issue or contact me directly for further inquiries.

---

Citation

If you use this code, please cite:

Rittmo, J., Franzmeier, N., Strandberg, O., Chauveau, L., Satterthwaite, T. D., Wisse, L. E., Spotorno, N., Behjat, H. H., Dehsarvi, A., Westen, D. van, Anijärv, T. E., Palmqvist, S., Janelidze, S., Stomrud, E., Ossenkoppele, R., Mattsson-Carlgren, N., Hansson, O., & Vogel, J. W. (2025). Age and Alzheimer’s disease affect functional connectivity along separate axes of functional brain organization (p. 2025.05.22.655469). bioRxiv. https://doi.org/10.1101/2025.05.22.655469

---