A Reproducibility Crisis for Clinical Metabolomics Studies

Authors

• Darcy Cochran, University of Nebraska-Lincoln
• Mai NourEldein, University of Nebraska-Lincoln
• Dominika Bezdekova, University of Nebraska-Lincoln
• Aaron Schram, University of Nebraska-Lincoln
• Réka Howard, University of Nebraska-Lincoln
• Robert Powers, University of Nebraska-Lincoln

Document Type

Article

Date of this Version

2024

Citation

Trends in Analytical Chemistry (2024) 180: 117918

doi: 10.1016/j.trac.2024.117918

Comments

Used by permission

Abstract

Cancer is a leading cause of world-wide death and a major subject of clinical studies focused on the identification of new diagnostic tools. An in-depth meta-analysis of 244 clinical metabolomics studies of human serum samples highlights a reproducibility crisis. A total of 2,206 unique metabolites were reported as statistically significant across the 244 studies, but 72% (1,582) of these metabolites were identified by only one study. Further analysis shows a random disparate disagreement in reported directions of metabolite concentration changes when detected by multiple studies. Statistical models revealed that 1,867 of the 2,206 metabolites (85%) are simply statistical noise. Only 3 to 12% of these metabolites reach the threshold of statistical significance for a specific cancer type. Our findings demonstrate the absence of a detectable metabolic response to cancer and provide evidence of a serious need by the metabolomics community to establish widely accepted best practices to improve future outcomes.