kaurao / eigenmodes_nulls

Geometric eigenmode null modeling

Code repository looking at geometric eigenmodes, and running some null modeling tests, in response to Pang et al. 2023, Nature.

External code required, including the openly available code release by Pang et al. and the BrainSpace toolbox. Both are included as sub-repos here. Other external code can be found in fcn/external. Special shoutout to numerical tours for great mesh processing tutorials and info.

External data required -- checkout the folder osf_dl for the link to the data.

Using surface-based geometric eigenmodes to explain randomized contrast maps

Figure 1 shows the results of the "spin test", in which the empirical surface-based eigenmodes from Pang et al were used to predict randomly rotated constrat maps. The code to reproduce this analysis can be found in run_spin_null.m. Figure 1 is based on 5000 random rotations of the constrat maps (nperms = 5e3). Generating these many rotations can be time consuming, and we recommend setting nperms = 1e2 for faster results.

Evaluating the reconstruction accuracy of geometric eigenmodes derived from other surfaces

Figure 2 uses geoemtric eigenmodes computed from an array of different surfaces to predict empirical constrat maps. Surfaces include a sphere as well as the very inflated, pial, midthickness, and white matter cortical meshes. In addition, we also consider ensembles of randomly perturbed spherical meshes. The steps to reproduce the results of Figure 2 are:

1. Run make_random_surfs.m to generate randomly perturbed spherical meshes.
2. Run eigenmode_calculation_othersurfs.sh and eigenmode_calculation_randsurfs.sh to compute the geometric eigenmodes for the array of different surfaces. This step is time consuming (we recommend running it on a high-performance computing platform) and requires the LaPy Python package.
3. Run run_alt_eigenmodes_recon.m and run_randsurfs_eigenmodes_recon.m.

and for additional analyses checkout the files in extra_scripts.