DynamGeom

MATLAB code for Stoll, Valluru & Rudebeck (2026) Dynamic geometry remapping of neural activity within frontal and subcortical areas during decision-making.

The behavioral and neurophysiological data are available on Zenodo (doi:10.5281/zenodo.17524410) and are described in London et al., Scientific Data (2026).

Requirements

• MATLAB R2024a (or newer) with the Statistics and Machine Learning Toolbox.
• Estimated marginal means for mixed-effects models (emmeans) toolbox from John Hartman (available on Github), bundled in scripts/emmeans/

Folder layout

• data_final/: per-session data (*_spk.mat, *_EOG.mat) and the pooled *_pool.mat files built by main_000_create_dataset.m.
• processed/: intermediate results that scripts pass to each other (see the table below).
• scripts/: main_*.m, utils/ (helper functions), emmeans/.
• report_XXX/: figures and a statistics log (output_XXX.txt) written by the corresponding script.

Pipeline

Scripts are numbered in the order they must be run. Each one reuses its cached processed/ files when present; set the overwrite* flags at the top of a script to recompute them.

Script	Analysis	Figures	Report folder	Reads (processed/)	Writes (processed/)
main_000_create_dataset.m	Pools raw spikes into data_final/*_pool.mat	–	–	–	–
main_000_behav.m	Choice behavior, saccades, spiking statistics	1B–F, S3	report_000	–	behav_pref.mat, saccadeCounts_*.mat, spiking_info.mat
main_001_anova_lda.m	Single-neuron ANOVAs, within-task LDA, task-general subspace	S1, S2	report_001	–	anova_lda.mat, flavor_1fc.mat, side_1fc.mat
main_002_states.m	Cross-task probability states, gaze-linked posteriors, area ablation	2B–F, 2H, S5A–D	report_002	behav_pref.mat, saccadeCounts_*.mat	states_2afc_final.mat
main_003_crossdecoding.m	2x2 cross-state decoding of chosen flavor and response side	3C–D, S6–S8	report_003	states_2afc_final.mat, behav_pref.mat	states_2afc_fr.mat, states_2afc_ccgp.mat
main_004_unitstability.m	Single-neuron tuning across probability states	S11	report_004	states_2afc_final.mat, flavor_1fc.mat, side_1fc.mat, states_2afc_ccgp.mat	–
main_005_thr_sensitivity.m	Robustness to state threshold and duration	S5E, S9	report_005 (one subfolder per threshold)	states_2afc_final.mat, saccadeCounts_reduced.mat, behav_pref.mat	states_2afc_fr_<thr>.mat, states_2afc_ccgp_<thr>.mat, states_2afc_thr_summary.mat
main_006_lda_vs_linear.m	Categorical (LDA) vs continuous (linear) probability decoders	S4	report_006	states_2afc_final.mat	–
main_007_statespace.m	Cross-validated 2D projections on chosen/unchosen-state LDA axes	S10	report_007	states_2afc_fr.mat, behav_pref.mat	states_2afc_cvplane.mat

states_2afc_final.mat holds the probability-state decoding of every session (built by main_002_states.m using the function scripts\utils\utils_decoding_crosstask_rmvarea.m). Every later file in the table derives from it: after rebuilding it, delete states_2afc_fr*.mat and states_2afc_ccgp*.mat (or set overwrite_fr / overwrite_decoding in main_003) so that main_003 and main_005 recompute them.

Usage

1. Place the data in data_final/ and run scripts/main_000_create_dataset.m once if the *_pool.mat files are missing.
2. In MATLAB, run run('scripts/run_all.m') from the project root.

run_all.m deletes the report folders of the main_* scripts before running them. Report files are overwritten on each run.

run_all.m also resets the random number generator (rng(seed, 'twister'), with seed = 55555 set at the top of run_all.m) before each script, so rebuilding the processed/ files reproduces the same pseudo-populations, cross-validation folds, trial subsamples and bootstraps. When running a script on its own, call rng(55555, 'twister') first to get the same results.

Scripts can also be run individually or section by section: each one resolves the project root from its own location (mfilename('fullpath')), or from the current folder when run section by section, so they do not depend on the MATLAB working directory.