RF1-SRA Trust

This is the authoritative downstream repository for RF1-SRA Trust Game fMRI analysis. It converts canonical BIDS events to FSL EVs, runs the established model-1 activation/seed-PPI/network-PPI models, and combines Trust runs 1 and 2 with fixed effects.

Reproducibility boundary

Production input comes only from rf1-sra-linux2:

• canonical task-trust BIDS events;
• fMRIPrep MNI-space BOLD images;
• TEDANA-enhanced FSL confounds.

This repository does not own raw logs, DICOM conversion, BIDS construction, fMRIPrep, TEDANA, or confound construction. Generated EVs and FEAT outputs live under ignored derivatives/fsl/. The frozen public teaching workflow uses OpenNeuro ds005123, version 1.1.3.

Pipeline

Linux2 canonical events ──> FSL three-column EVs ──> L1 model-1
Linux2 fMRIPrep BOLD ──────────────────────────────> activation / seed PPI / nPPI
Linux2 TEDANA confounds ───────────────────────────> runs 1 + 2 ──> L2 fixed effects

The established activation model has 10 task EVs and 18 contrasts. Seed PPI and DMN/ECN nPPI each retain 19 contrasts. See templates/README.md for the exact model contract.

Internal quick start

On Linux2:

cd /ZPOOL/data/projects/rf1-sra-trust
git pull --ff-only origin main
bash code/validate_workflow.sh

python3 code/build_L1_manifest.py \
  --output logs/runlists/L1-ready.tsv \
  --missing-output logs/runlists/L1-missing.tsv

bash code/run_logged.sh --label trust-EVs -- \
  bash code/run_gen3colfiles.sh \
    --manifest logs/runlists/L1-ready.tsv --jobs 8

bash code/run_logged.sh --label trust-L1-activation -- \
  bash code/run_L1stats.sh \
    --manifest logs/runlists/L1-ready.tsv --ppi 0 --jobs 50 \
    --log-dir logs/L1-activation-current \
  --check python3 code/audit_outputs.py \
    --level l1 --manifest logs/runlists/L1-ready.tsv --type act \
    --output logs/records/L1-act-completeness.tsv

Build L2 readiness only after both L1 runs are complete:

python3 code/build_L2_manifest.py --type act \
  --output logs/runlists/L2-act-ready.tsv \
  --missing-output logs/runlists/L2-act-missing.tsv

bash code/run_logged.sh --label trust-L2-activation -- \
  bash code/run_L2stats.sh \
    --manifest logs/runlists/L2-act-ready.tsv --type act --jobs 20 \
    --log-dir logs/L2-activation-current \
  --check python3 code/audit_outputs.py \
    --level l2 --manifest logs/runlists/L2-act-ready.tsv --type act \
    --output logs/records/L2-act-completeness.tsv

For connectivity, run activation first, then use --ppi VS, --ppi dmn, or --ppi ecn. Build the matching L2 manifest with --type ppi_seed-VS, nppi-dmn, or nppi-ecn.

To launch activation and one seed-PPI model together without doubling concurrency, use code/run_L1_activation_ppi.sh. Each run-level worker completes activation before its matching PPI model; --jobs 50 therefore means at most 50 simultaneous FEAT processes total.

For paired fixed effects, code/run_L2_activation_ppi.sh provides the same sequencing at the subject-session level. --jobs 20 means at most 20 concurrent L2 models, while L2stats.sh keeps within-model FSL submission serial.

Public teaching quick start

Open notebooks/README.md and run notebooks 01 → 02 → 03 in Neurodesk. They download only one public participant’s Trust files and call the same production scripts/templates used above.

Repository layout

• code/: active manifest, EV, L1/L2, QC, completeness-audit, validation, and logging tools; historical material is under code/archive/.
• templates/: sole active model-1 FEAT templates; historical L3 materials are archived.
• masks/: the retained VS seed and Smith-network maps with provenance/geometry notes.
• notebooks/: public Neurodesk teaching workflow.
• tests/: lightweight scientific and path-contract tests.
• logs/records/: compact Git-trackable run records; raw logs remain ignored.

Historical notes

This repository originated as an SRNDNA Trust repository and previously mixed preprocessing, behavioral conversion, analysis, extraction, and presentation assets. The current workflow is intentionally narrower. Historical behavioral/support code and L3 material remain clearly archived and are not production-supported. See code/WORKFLOW_AUDIT.md.