AUTHOR=Li Xin , Zhao Xiaoyue , Zhang Ruonan , Zhuang Xuewei 

TITLE=Harnessing serum VOCs and machine learning for the early detection of MAFLD

JOURNAL=Frontiers in Endocrinology

VOLUME=Volume 16 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/endocrinology/articles/10.3389/fendo.2025.1691853

DOI=10.3389/fendo.2025.1691853

ISSN=1664-2392

ABSTRACT=IntroductionMetabolic dysfunction-associated fatty liver disease (MAFLD) is a complex metabolic disorder and one of the leading causes of chronic liver disease worldwide. Current diagnostic tools, such as ultrasound, lack sufficient sensitivity for detecting early-stage disease, emphasizing the urgent need for novel and non-invasive diagnostic strategies. Metabolomics, particularly the profiling of volatile organic compounds (VOCs) in biofluids, has emerged as a promising approach for biomarker discovery in metabolic diseases.MethodsIn this preliminary single-center study, serum samples were collected from 199 participants, including 110 MAFLD patients and 89 healthy controls. Volatile organic compounds were analyzed using gas chromatography–ion mobility spectrometry (GC-IMS). Machine learning algorithms, including random forest, were applied to construct diagnostic models and identify key discriminatory metabolites. Clinical and biochemical parameters such as age, body mass index, liver function, and lipid profiles were also compared between groups.ResultsA total of 79 serum VOCs were detected, among which 54 showed significant differences between MAFLD patients and controls (29 identified and 25 unidentified). The random forest model exhibited the best diagnostic performance, achieving a test AUC of 0.941, with 86.7% sensitivity and 88.5% specificity. Seven key VOCs were identified as important contributors to the model, including two upregulated compounds (2-Butoxyethanol and Cyclopentanone-D) and five downregulated compounds ((E)-3-hexenoic acid, 2-Ethylbutanal, 2-Propyl acetate, Benzaldehyde-M, and Furaneol). Notably, 2-pentylfuran displayed significant variation across different pathological grades of MAFLD, suggesting potential as a stage-specific biomarker.DiscussionThis study demonstrates that serum VOC profiling using GC-IMS combined with machine learning can effectively distinguish MAFLD patients from healthy individuals. The identified VOC signatures, particularly 2-pentylfuran, may serve as non-invasive biomarkers for MAFLD diagnosis and staging. However, due to the limited sample size and single-center design, these findings require validation in larger, multi-center, and longitudinal studies to confirm their clinical applicability, especially for early disease detection.