Machine learning classification of fibromyalgia from blood transcriptomes: confounder benchmarking and implications for endometriosis-associated pain
- 1Faculty of Medicine of Ibn El Jazzar, Sousse
Abstract
Fibromyalgia (FM) lacks an objective biomarker, and blood transcriptomic signatures have been proposed to fill this gap. Here, the largest public peripheral blood mononuclear cell (PBMC) RNA-seq cohort of FM reporting expression subtypes (96 patients, 93 controls) was reanalysed using leakage-free nested cross-validation, with feature selection and scaling confined strictly to training folds. Gene expression discriminated patients from controls (ROC-AUC 0.821 for logistic regression and 0.825 for XGBoost), clearing a 100-repetition permutation-label null (p= 0.0099). Two simple, disease-agnostic baselines, however, reach nearly the same performance: sex alone reaches AUC 0.727, and a four-score PBMC cell-composition proxy reaches AUC 0.71-0.76. Restricting to females alone gives AUC 0.783 (0.705-0.858), and excluding X/Y-linked genes does not reduce performance, so the signal is not purely a sex-composition artifact; but the female-only expression model does not significantly exceed the composition-only proxy in the same females (0.783 vs. 0.712, paired 95% CI -0.025 to 0.175), so a composition contribution cannot be excluded. Preranked gene set enrichment analysis finds curated CLEAR/lysosomal and interferon/JAK-STAT gene sets, proxies for two subtype signatures of the original study, enriched among genes expressed lower in FM, also in females alone, but this enrichment disappears after adjusting for the composition proxies. Gynecological comorbidity, hormonal status, and medication are absent from the public metadata and remain plausible, untested contributors to the female-only estimate. These findings are hypothesis-generating, not a basis for clinical diagnosis: a single-cohort, correlational result consistent with, but not sufficient to establish, an FM-specific signal distinct from demographic, hormonal, and cell-composition confounds, pending independent replication. Future blood-based biomarker studies for FM, and for related nociplastic pain conditions such as endometriosis, should report performance relative to sex- and composition-based baselines as standard practice.
Related
- GGE: General-purpose deep meta-learning for classification of human transcriptomes with limited data
- Quantifying the reliability gap in cross-domain plant disease classification: benchmarking the limited efficacy of standard mitigation techniques under controlled-to-field shift
- A claim-centred atlas distinguishes biological corroboration from analytical depth in plant stress transcriptomes
- Preoptic transcriptomes during courtship reveal the molecular basis of behavioral coordination in a highly social cichlid
- PatHoST: weakly supervised reconstruction of pathogen spatial distributions from host transcriptomes
BibTeX
@misc{ouni2026machine,
title = {{Machine learning classification of fibromyalgia from blood transcriptomes: confounder benchmarking and implications for endometriosis-associated pain}},
author = {S. Ouni and F. Ouni and A. Chebbi},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.21.753755},
url = {https://www.biorxiv.org/content/10.64898/2026.10.21.753755v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Machine learning classification of fibromyalgia from blood transcriptomes: confounder benchmarking and implications for endometriosis-associated pain
AU - S. Ouni
AU - F. Ouni
AU - A. Chebbi
PY - 2026
DA - 2026/10/21
PB - bioRxiv
DO - 10.64898/2026.10.21.753755
UR - https://www.biorxiv.org/content/10.64898/2026.10.21.753755v1
AB - Fibromyalgia (FM) lacks an objective biomarker, and blood transcriptomic signatures have been proposed to fill this gap. Here, the largest public peripheral blood mononuclear cell (PBMC) RNA-seq cohort of FM reporting expression subtypes (96 patients, 93 controls) was reanalysed using leakage-free nested cross-validation, with feature selection and scaling confined strictly to training folds. Gene expression discriminated patients from controls (ROC-AUC 0.821 for logistic regression and 0.825 for XGBoost), clearing a 100-repetition permutation-label null (p= 0.0099). Two simple, disease-agnostic baselines, however, reach nearly the same performance: sex alone reaches AUC 0.727, and a four-score PBMC cell-composition proxy reaches AUC 0.71-0.76. Restricting to females alone gives AUC 0.783 (0.705-0.858), and excluding X/Y-linked genes does not reduce performance, so the signal is not purely a sex-composition artifact; but the female-only expression model does not significantly exceed the composition-only proxy in the same females (0.783 vs. 0.712, paired 95% CI -0.025 to 0.175), so a composition contribution cannot be excluded. Preranked gene set enrichment analysis finds curated CLEAR/lysosomal and interferon/JAK-STAT gene sets, proxies for two subtype signatures of the original study, enriched among genes expressed lower in FM, also in females alone, but this enrichment disappears after adjusting for the composition proxies. Gynecological comorbidity, hormonal status, and medication are absent from the public metadata and remain plausible, untested contributors to the female-only estimate. These findings are hypothesis-generating, not a basis for clinical diagnosis: a single-cohort, correlational result consistent with, but not sufficient to establish, an FM-specific signal distinct from demographic, hormonal, and cell-composition confounds, pending independent replication. Future blood-based biomarker studies for FM, and for related nociplastic pain conditions such as endometriosis, should report performance relative to sex- and composition-based baselines as standard practice.
ER -