Background <p>Cholangiocarcinoma (CCA) is a rare and highly aggressive malignancy originating in the bile ducts. Owing to limitations involving pathological sampling, the clinical differentiation of CCA from benign biliary diseases remains challenging. This study aimed to evaluate the differences between the bile lipidomes of CCA patients and those of patients with benign disease to develop a bile lipid classifier that can help to differentiate CCA from benign conditions.</p> Methods <p>Bile samples were collected by endoscopic retrograde cholangiography (ERCP) from patients with CCA or benign disease. The participants were divided into three cohorts: the first two cohorts underwent untargeted lipidomic analysis, whereas the third cohort was subjected to targeted lipid quantification. Untargeted lipidomic analysis was performed via ultrahigh-performance liquid chromatography coupled with ion mobility quadrupole time-of-flight mass spectrometry (UHPLC/IM-QTOF-MS). Targeted lipid quantification was conducted via UHPLC‒MS/MS in multiple reaction monitoring (MRM) mode. Lipid features were screened to construct a bile lipid classifier using the machine learning algorithm, least absolute shrinkage and selection operator (LASSO) regression, followed by cross-validation in two cohorts. The selected lipid features were further validated by targeted quantification in the third cohort. The functions of the significantly differentially abundant lipids in proliferation were validated in CCA cell lines.</p> Results <p>In total, 241 bile samples were collected and divided into three cohorts for independent lipidomic analysis: Cohort 1 included 32 CCA samples and 68 benign controls; Cohort 2 included 30 CCA samples and 30 benign controls; and Cohort 3 included 32 CCA samples and 49 benign controls. There were significant differences in the lipid profiles of the bile samples obtained from patients with CCA and individuals with benign disease, with multiple lipid classes, particularly lysophosphatidylcholine (LPC), significantly downregulated in the CCA group. Multimodule correlation networks constructed via weighted lipid coexpression network analysis (WLCNA) revealed significant associations between lipid modules and clinical traits. A machine learning-based bile lipid classifier, termed BileLipid, was developed for CCA diagnosis; this classifier incorporates six lipid features. This classifier achieved areas under the receiver operating characteristic curve (AUCs) of 0.943, 0.956, and 0.828 in Cohorts 1, 2, and 3, respectively. Additionally, the significantly downregulated lipid LPC in CCA bile was found to significantly inhibit the proliferation of CCA cell lines, suggesting its potential role as a protective factor in CCA.</p> Conclusions <p>This study not only identified lipidomic alterations in CCA using bile samples but also established and validated a bile lipid classifier with high specificity and sensitivity for distinguishing between CCA and benign bile duct diseases. Our findings highlight the potential of bile lipid biomarkers for improving the differential diagnosis and risk assessment of CCA and preventing potential overintervention in patients with benign biliary disease.</p>