Identification and validation of volatile organic compounds in bile for differential diagnosis of perihilar cholangiocarcinoma

Early and differential diagnosis of perihilar cholangiocarcinoma (PHCCA) is highly challenging. This study aimed to evaluate whether volatile organic compounds (VOCs) in bile samples could be emerging diagnostic biomarkers for PHCCA. We collected 200 bile samples from patients with PHCCA and benign biliary diseases (BBD), including a 140-patient training cohort and an 60-patient test cohort. Gas chromatography-ion mobility spec-trometry (GC-IMS) was used for VOCs detection. The predictive models were constructed using machine learning algorithms. Our analysis detected 19 VOC substances using GC-IMS in the bile samples and resulted in the identification of three new VOCs, 2-methoxyfuran, propyl isovalerate, and diethyl malonate that were found in bile. Unsupervised hierarchical clustering analysis supported that VOCs detected in the bile could distinguish PHCCA from BBD. Twelve VOCs defined according to 32 signal peaks had significant statistical significance between BBD and PHCCA, including four up-regulated VOCs in PHCCA, such as 2-ethyl-1-hexanol, propyl iso-valerate, cyclohexanone, and acetophenone, while the rest eight VOCs were down-regulated. ROC curve analysis revealed that machine learning models based on VOCs could help diagnosing PHCCA. Among them, SVM pro-vided the highest AUC of 0.966, with a sensitivity and specificity of 93.1% and 100%, respectively. The diag-nostic model based on different VOC spectra could be a feasible method for the differential diagnosis of PHCCA.

Automated summary

This study aimed to evaluate whether volatile organic compounds (VOCs) in bile samples could be emerging diagnostic biomarkers for perihilar cholangiocarcinoma (PHCCA). Gas chromatography-ion mobility spectrometry (GC-IMS) was used to detect 19 VOC substances in the bile samples and identified three new VOCs. Cluster analysis supported that VOCs detected in the bile could distinguish PHCCA from benign biliary diseases (BBD). Machine learning models based on VOCs showed high diagnostic accuracy for PHCCA, with the support vector machine (SVM) providing the highest area under the curve (AUC) value of 0.966.