PreprintCC BY-NC-NDmedRxiv
Performance of protein panels is inflated across many biomarker studies
- 1Lund University
Abstract
Data leakage is a prevalent yet underappreciated flaw in biomarker discovery studies. Through simulation and real-world proteomic data, we demonstrate that typical pipelines are broadly susceptible to this issue, producing inflated performance estimates, poor generalization, and excess false positives. We further introduce two tools to detect data leakage at the code and manuscript level, providing a practical path toward more rigorous and reproducible biomarker reporting.
Related
- Host-response biomarker panels for distinguishing selected acute respiratory viral-bacterial co-infections from mono-infections
- Circulating and extracellular vesicle-associated non-coding RNAs in diabetic retinopathy: a stage-aware framework for network-informed biomarker panels
- Finding Safety Signals in Sparse Data: Zero-Inflated Models for Pharmacovigilance studies
- Confirmatory biomarker diagnostic studies are not needed when transitioning from NAFLD to MASLD
- CHCHD2 as a prognostic biomarker and therapeutic target in lung adenocarcinoma: Insights from pan-cancer multi-omics and functional studies
BibTeX
@misc{an2026performance,
title = {{Performance of protein panels is inflated across many biomarker studies}},
author = {L. An and C. A. Finney and A. Shvetcov and The Global Neurodegeneration Proteomics Consortium and J. Vogel},
howpublished = {medRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.02.26362037},
url = {https://www.medrxiv.org/content/10.64898/2026.09.02.26362037v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Performance of protein panels is inflated across many biomarker studies
AU - L. An
AU - C. A. Finney
AU - A. Shvetcov
AU - The Global Neurodegeneration Proteomics Consortium
AU - J. Vogel
PY - 2026
DA - 2026/09/02
PB - medRxiv
DO - 10.64898/2026.09.02.26362037
UR - https://www.medrxiv.org/content/10.64898/2026.09.02.26362037v1
AB - Data leakage is a prevalent yet underappreciated flaw in biomarker discovery studies. Through simulation and real-world proteomic data, we demonstrate that typical pipelines are broadly susceptible to this issue, producing inflated performance estimates, poor generalization, and excess false positives. We further introduce two tools to detect data leakage at the code and manuscript level, providing a practical path toward more rigorous and reproducible biomarker reporting.
ER -