Skip to content

REPRODUCE.md

Latest commit

 

History

History
41 lines (35 loc) · 2.37 KB

REPRODUCE.md

File metadata and controls

41 lines (35 loc) · 2.37 KB
Raw

Reproduce — SPaRCNet (Jing et al., Neurology 2023)

Every main-text figure/table regenerates from committed, de-identified proximal data under code_for_figures/Data/ (surrogate subject IDs, e.g. sparcnet_subject1026; no PHI). No raw data download or model retraining needed.

One command

cd code_for_figures
run_all           % see run_all.m — runs every script below, writes the PNGs

Or run any single script (each cd-independent from code_for_figures/):

Paper item Script Input (committed) Output
Table 1 (cohort splits) Table1_Splits.m Data/Table1/patient_demo.mat, Data/Table1/dataset{1..4}.mat stdout
Figure 1 (ROC) Figure1_ROC.m Data/Figure1/figure1_input.mat Fig1.png
Figure 2 (PR) Figure2_PR.m Data/Figure1/figure1_input.mat Fig2.png
Figure 3 (UMAP "starfish") Figure3_UMAPs.m Data/Figure3/figure3_input.mat, Data/Figure3/samples.mat Fig3.png
Figure 4 (SZ/LPD samples) Figure4_samples_SZ_LPD.m Data/Figure3/samples.mat Fig4.png
Figure 5 (GPD/LRDA samples) Figure5_samples_GPD_LRDA.m Data/Figure3/samples.mat Fig5.png
Figure 6 (GRDA/Other samples) Figure6_samples_GRDA_Other.m Data/Figure3/samples.mat Fig6.png
Figure S2 (flowchart) FigureS2_Flowchart.m Data/FigureS2/*.mat figure
Figure S3 (SP spread) FigureS3_SPspread.m Data/FigureS3/FigureS3_input.mat FigS3.png
Figure S5 (UMAP spread) FigureS5_UMAPspread.m Data/FigureS5/FigureS5_input.mat FigS5.png
Figure S8 (IRR) FigureS8_IRR.m Data/FigureS8/FigureS8_input.mat FigS8.png

Committed reference outputs (Fig1.png … FigS8.png) let you diff your regenerated figures against the published versions.

Requirements

  • MATLAB R2019b+ (developed on R2020a). Toolboxes: Statistics and Machine Learning Toolbox. Figure 3/S5 use UMAP — the umap File Exchange package must be on the MATLAB path (the 2-D coordinates are precomputed in figure3_input.mat, so UMAP is only needed if you recompute the embedding).

Provenance

The committed .mat are the proximal artifacts (model outputs + expert labels + 2-D embeddings) derived from the raw SPaRCNet EEG dataset. See DATA_SOURCE.md for the raw-data home and the raw→derived lineage.