HassanB4/warfarisk-autogluon-1780

Beyond MAE: Phase 3 AutoGluon (IWPC-1780)

This repository contains the Phase 3 winning model from the pipeline described in "Beyond Mean Absolute Error: Ancestry-Stratified Calibration and Explainability for Warfarin Dosing Models", a nine-phase reproducible ML pipeline for warfarin dose prediction, trained on the IWPC-1780 cohort. An AutoGluon extreme_quality predictor on the combined clinical+genetic feature set reaches MAE 7.959, R² 0.480, PW20 0.497, the best of 9 architectures × 3 feature sets evaluated on this cohort, with the stacking ensemble (R² 0.479) essentially tied for second.

By: Hassan Barmandah, Omar Abdullah Bawazir, Siraj Aldeen Marghalani, Moath Shaat, Abdullah N. Alkattan, and Mariam M. AlEissa (corresponding author), Alfaisal University, Riyadh (with Umm Al-Qura University, Saudi Electronic University, Ministry of Health, Public Health Authority, King Khaled Eye Specialist Hospital Research Center, King Abdullah Petroleum Studies and Research Center)

This is a research artifact, not a validated clinical tool. It has not been evaluated prospectively and has no regulatory status.

---

Model Description

This model is an AutoGluon TabularPredictor fit with the extreme_quality preset on IWPC-1780 (n=1,780, n_test=356, fixed 80/20 patient-level split, seed 20260725), using the combined feature set (clinical + CYP2C9/VKORC1 genotype consensus). IWPC-1780 uses binary ancestry flags (Black, Asian) and pre-z-scored continuous features, and is evaluated independently rather than pooled with IWPC-6256. This is the best-performing model of 9 architectures evaluated across 3 feature sets on this cohort in the Phase 3 nested ablation.

Key Contributions

• Best Single Model on IWPC-1780: MAE 7.959 / R² 0.480 / PW20 0.497, narrowly ahead of the stacking ensemble (R² 0.479) and TabPFN (R² 0.477), notably beating every individually-tuned model including CatBoost/TabPFN on this smaller cohort
• Combined-Feature Advantage: beats the same architecture family's clinical-only (R² 0.296) and genetic-only (R² 0.313) results on this cohort
• Independent Second-Cohort Replication: run with the identical AutoGluon configuration used on IWPC-6256, confirming the combined-feature advantage is not an artifact of one dataset

---

🚀 How to Use

from autogluon.tabular import TabularPredictor
from huggingface_hub import snapshot_download

local_dir = snapshot_download("HassanB4/warfarisk-autogluon-1780")
predictor = TabularPredictor.load(local_dir)
prediction = predictor.predict(your_dataframe)  # combined clinical+genetic features, see Training Data below

Input features: age, weight, height (pre-z-scored, not raw units; do not feed raw clinical values without re-harmonizing scale), binary ancestry flags (Black, Asian), CYP2C9 consensus, VKORC1.AA/VKORC1.AG genotype flags, amiodarone/enzyme-inducer comedication flags. See HasanBGit/WarfaRisk's src/warfarisk/phase3_genetics_ablation.py for the exact feature-engineering pipeline.

---

⚙️ Training Procedure

Training Data

Split	Samples	Description
Training	1,424	80% of IWPC-1780, patient-ID-level split
Test	356	Held out, seed 20260725

Hyperparameters

Parameter	Value	Parameter	Value
Preset	extreme_quality	Time Limit	600s
Feature Set	combined (clinical + genetic)	Excluded Model Types	TABDPT, TABICL, TABM, MITRA
Split Seed	20260725	Split Level	Patient ID (not row)
Final Ensemble	WeightedEnsemble_L2 (blend of RealTabPFN-v2_r13/LightGBMPrep_r13/RealTabPFN-v2_r106)	Leakage Audit	11/11 checks passed

Same excluded-model-type configuration as the IWPC-6256 run, applied for consistency across cohorts. Full run notes in docs/RESULTS.md and docs/KNOWN_ISSUES.md.

Models Compared (combined feature set, IWPC-1780)

Model	MAE	R²	PW20
AutoGluon extreme_quality	7.959	0.480	0.497
Stacking Ensemble	7.993	0.479	0.514
TabPFN	8.018	0.477	0.483
Linear Regression	8.070	0.474	0.511
CatBoost	8.205	0.451	0.492
Random Forest	8.940	0.370	0.444
MLP	9.231	0.320	0.447
Elastic Net	9.404	0.286	0.449
XGBoost	10.583	0.165	0.396

---

📊 Evaluation Results

Metric	Value
MAE (mg/week)	7.959
R²	0.480
PW20	0.497

PW20 = proportion of predictions within 20% of the true dose (IWPC's own clinical-acceptability threshold is ≥0.50: this model comes within one point of clearing it, the closest of any model evaluated in this project).

Comparison Across Cohorts

Cohort	MAE	R²	PW20
IWPC-6256 (phase3-autogluon-iwpc6256)	8.552	0.475	0.466
IWPC-1780 (this model)	7.959	0.480	0.497

---

⚠️ Limitations

• Research-Only: not validated prospectively, no regulatory status
• Pre-Scaled Features: this cohort's continuous features (age/weight/height) are pre-z-scored at source; do not pool raw inputs with IWPC-6256 without re-harmonizing scale
• Coarser Ancestry Encoding: IWPC-1780 only distinguishes Black/Asian/Other, unlike IWPC-6256's 14 labels; see docs/RESULTS.md for this project's Phase 4/5 ancestry-fairness and calibration findings on this cohort
• No Raw Data Shipped: the code repository does not redistribute IWPC data; see data/DATA.md

---

🙏 Acknowledgements

We thank the PharmGKB / International Warfarin Pharmacogenetics Consortium for the IWPC dataset, and the AutoGluon team for the AutoML framework this model was built with.

Related Links

• AutoGluon
• PharmGKB
• Code repository

---

📜 Citation

This model is described in the following manuscript, submitted to the MDPI journal AI and under review as of August 2026. The DOI below will be updated once the paper is formally published.

Barmandah, H.; Bawazir, O.A.; Marghalani, S.A.; Shaat, M.; Alkattan, A.N.; AlEissa, M.M. Beyond Mean Absolute Error: Ancestry-Stratified Calibration and Explainability for Warfarin Dosing Models. AI 2026, submitted.

@article{barmandah2026beyond,
    title={Beyond Mean Absolute Error: Ancestry-Stratified Calibration and Explainability for Warfarin Dosing Models},
    author={Barmandah, Hassan and Bawazir, Omar Abdullah and Marghalani, Siraj Aldeen and Shaat, Moath and Alkattan, Abdullah N. and AlEissa, Mariam M.},
    journal={AI},
    publisher={MDPI},
    year={2026},
    note={Manuscript submitted, under review as of August 2026. Cite the published DOI once assigned.}
}

---

📄 License

This project is licensed under the Apache 2.0 License. This model was fit on IWPC data; the fitted artifact and code are shared under Apache-2.0, but the underlying IWPC dataset is not redistributed by this repository; see HasanBGit/WarfaRisk's data/DATA.md for how to obtain it.