AUTHOR=Qian Jingrong , Liu Qi , Wang Jue , Zhuang Xuewei , Fang Jun 

TITLE=Identifying novel biomarkers for biliary tract cancer based on volatile organic compounds analysis and machine learning

JOURNAL=Frontiers in Oncology

VOLUME=Volume 15 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2025.1572460

DOI=10.3389/fonc.2025.1572460

ISSN=2234-943X

ABSTRACT=BackgroundThe current diagnostic methods for biliary tract cancer (BTC) have limitations in sensitivity and specificity. This study aims to explore the use of volatile organic compounds (VOCs) in serum to distinguish BTC and benign biliary diseases (BBD).MethodWe collected 158 serum samples from BTC and BBD patients, and used gas chromatography ion mobility spectrometry (GC-IMS) for VOCs detection. Six machine learning methods (RF, SVM, LDA, KNN, LASSO, and XGBoost) were used to construct and evaluate diagnostic prediction models.ResultWe detected a total of 40 VOCs in patients, of which 14 VOCs were statistically significant (p < 0.05), including 11 up-regulated and 3 down-regulated VOCs. In BTC and BBD patients, the diagnostic model was constructed based on six machine learning method. Among them, RF had the highest diagnostic performance (AUC = 0.935, p < 0.001), with a sensitivity of 76.2% and a specificity of 96.3%. Based on the importance score, we selected the top 4 VOCs, and constructed an optimized diagnostic model through five fold cross validation. The model’s AUC was 0.982, sensitivity was 87.9%, and specificity was 96.7%, which improved the diagnostic sensitivity and reduced FNR. In addition, in patients with cholangiocarcinoma and BBD, we further screened for 4-VOCs and constructed diagnostic model, with an AUC of 0.977, accuracy of 92.4%, specificity of 98.9%, sensitivity of 77.5%.ConclusionThe diagnostic model based on 4-VOCs may be a feasible method for distinguishing the diagnosis of BTC and BBD patients.