Site-Level Distribution Shift Dominates Disease Signal and Undermines Cross-Site Generalization in High-Dimensional Microbiome Classifiers
Fuente:
Zenodo
Saved in:
| Main Author: | |
|---|---|
| Format: | Recurso digital |
| Published: |
Zenodo
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866901924074225664 |
|---|---|
| author | Pereira, Antonio |
| author_facet | Pereira, Antonio |
| contents | <p>This repository contains the complete reproducible analysis pipeline supporting the manuscript:</p> <p>“Site-Level Distribution Shift Dominates Disease Signal and Undermines Cross-Site Generalization in High-Dimensional Microbiome Classifiers.”</p> <p>The study evaluates cross-site portability of Parkinson’s disease (PD) microbiome classifiers across three geographically distinct cohorts (Finland, Malaysia, USA; n = 682).</p> <p>Key findings:</p> <p>• Strict within-cohort cross-validation yields high discrimination (AUC up to 0.944).<br>• Leave-one-cohort-out validation results in substantial degradation (AUC as low as 0.506).<br>• Cohort identity explains 68.9% of compositional variance (PERMANOVA).<br>• PD status explains 0.92%.<br>• Disease-associated effect vectors are nearly orthogonal to the dominant ecological axis (PC1 explaining 53.5% of total variance).</p> <p>The repository includes:</p> <p>• Harmonized genus-level count matrices (108 shared genera)<br>• Centered log-ratio transformed feature matrices<br>• Complete R analysis pipeline (scripts C21–C33)<br>• Within-cohort and leave-one-cohort-out performance outputs<br>• PCA structure analyses and PD effect alignment quantification<br>• All summary tables used in the manuscript</p> <p>All analyses were conducted in R 4.4.2 with deterministic random seeds.</p> <p>This version corresponds to the IEEE JBHI submission version (v1.0-portability).</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_18817750 |
| institution | Zenodo |
| language | |
| publishDate | 2026 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | Site-Level Distribution Shift Dominates Disease Signal and Undermines Cross-Site Generalization in High-Dimensional Microbiome Classifiers Pereira, Antonio distribution shift cross-site validation biomedical machine learning Parkinson Disease generalization compositional data analysis <p>This repository contains the complete reproducible analysis pipeline supporting the manuscript:</p> <p>“Site-Level Distribution Shift Dominates Disease Signal and Undermines Cross-Site Generalization in High-Dimensional Microbiome Classifiers.”</p> <p>The study evaluates cross-site portability of Parkinson’s disease (PD) microbiome classifiers across three geographically distinct cohorts (Finland, Malaysia, USA; n = 682).</p> <p>Key findings:</p> <p>• Strict within-cohort cross-validation yields high discrimination (AUC up to 0.944).<br>• Leave-one-cohort-out validation results in substantial degradation (AUC as low as 0.506).<br>• Cohort identity explains 68.9% of compositional variance (PERMANOVA).<br>• PD status explains 0.92%.<br>• Disease-associated effect vectors are nearly orthogonal to the dominant ecological axis (PC1 explaining 53.5% of total variance).</p> <p>The repository includes:</p> <p>• Harmonized genus-level count matrices (108 shared genera)<br>• Centered log-ratio transformed feature matrices<br>• Complete R analysis pipeline (scripts C21–C33)<br>• Within-cohort and leave-one-cohort-out performance outputs<br>• PCA structure analyses and PD effect alignment quantification<br>• All summary tables used in the manuscript</p> <p>All analyses were conducted in R 4.4.2 with deterministic random seeds.</p> <p>This version corresponds to the IEEE JBHI submission version (v1.0-portability).</p> |
| title | Site-Level Distribution Shift Dominates Disease Signal and Undermines Cross-Site Generalization in High-Dimensional Microbiome Classifiers |
| topic | distribution shift cross-site validation biomedical machine learning Parkinson Disease generalization compositional data analysis |
| url | https://doi.org/10.5281/zenodo.18817750 |