The IQ-Motion Confound in Multi-Site Autism fMRI May Be Inflated by Site-Correlated Measurement Uncertainty

Author: Kareem Soliman

This repository provides a self-contained replication of the ABIDE-I phenotypic analysis reported in the paper. It reimplements only the published equations needed to estimate the pooled OLS slope, the EIV-corrected PCR slope, the leave-site-out cross-validation result, the within-tier slope analysis, and the 8x8 sensitivity grid.

PCR is introduced as a novel method in this work. This repository contains the implementation of Probability Cloud Regression (PCR) sufficient to reproduce the reported results.

Setup

Python 3.9+ is sufficient.

pip install -r requirements.txt

Data

Download the public ABIDE-I phenotypic CSV separately, then place it in data/abide_phenotypic.csv. Detailed instructions are in data/README.md.

Usage

python src/run_analysis.py

If the CSV lives elsewhere:

python src/run_analysis.py --input /path/to/abide_phenotypic.csv

Expected Output

Running the full pipeline will:

• filter the ABIDE-I phenotypic CSV to the paper sample (n = 935, 19 sites)
• print the OLS fit, PCR fit, EM convergence trace, Table 2 tier slopes, and LOSO cross-validation summary
• save privacy-safe CSV outputs under results/ (no subject identifiers such as SUB_ID or FILE_ID)
• save four PNG figures under figures/
• run a verification block against the reported paper values

At baseline, the expected headline values are:

• OLS slope: about -0.00125
• PCR slope: about -0.00027
• bias factor: about 4.67x
• LOSO R^2: about -0.074

The manuscript reports convergence in 47 iterations for the original analysis run. This rounded-parameter reimplementation converges faster on the public CSV while still matching the reported headline results within tolerance.

Privacy and Data Distribution

• This repository does not distribute the raw ABIDE-I phenotypic CSV.
• Users must download ABIDE-I from the official source and follow its terms and conditions.
• Published outputs in this repo are reduced to reproducibility-focused, privacy-safe tables and omit direct subject identifiers.

License and Intended Use

• This repository is released for academic reproducibility and research use.
• Commercial use is not permitted under the license applied to this repository.
• If you need commercial licensing, contact the author.

Citation

If you use this repository, please cite:

@article{soliman2026iqmotion,
  title={The IQ-Motion Confound in Multi-Site Autism fMRI May Be Inflated
         by Site-Correlated Measurement Uncertainty},
  author={Soliman, Kareem},
  year={2026}
}

Contributing

Reproduction reports and bug fixes are welcome — see CONTRIBUTING.md.

Community and Governance

• Questions and replication discussion: GitHub Discussions
• Security reporting: SECURITY.md
• Community expectations: CODE_OF_CONDUCT.md

Contact

Kareem Soliman — Independent AI Researcher

• GitHub: @kareem-soliman-ai
• Email: kareem.soliman@outlook.com.au

Licence

CC BY-NC-SA 4.0 — see LICENSE for details.