2743 - Metabolomics and Machine Learning Identify Novel Serum Biomarkers for Nasopharyngeal Carcinoma Diagnosis
Presenter(s)
J. Yang1, S. Gao2, X. Yin3, J. Ding1, X. Qiu1, Z. Fei1, and J. Dong2; 1Department of Radiation Oncology, Clinical Oncology School of Fujian Medical University, Fujian Cancer Hospital, Fuzhou, Fujian, China, 2Xiamen University, Xiamen, Fujian, China, 3Shandong Cancer Hospital and Institute, Shandong First Medical University and Shandong Academy of Medical Sciences Department of Radiation Oncology, Jinan, Shandong, China
Purpose/Objective(s): This study aims identify metabolic features and biomarkers of NPC through metabolomics analysis combined with machine learning techniques to establish a model for NPC diagnosis.
Materials/Methods: Serum samples from 126 NPC patients and 50 controls across two independent clinical centers were analyzed via liquid chromatography-mass spectrometry (LC-MS) for non-targeted metabolomics. After data preprocessing, statistical analysis was performed and principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA) were applied for data downscaling and classification. We employed the Recursive Feature Elimination (REF) method to perform bidirectional feature selection for constructing a logistic regression model, followed by grid search optimization of model parameters. Machine learning algorithms, including support vector machine (SVM), random forest (RF), PLS-DA, and logistic regression(LR), were used to screen the metabolic markers which can distinguish the experimental group from the control, and to construct a NPC diagnostic model.
Results: A total of 4425 metabolites were detected by metabolomics, and PCA revealed altered metabolic profiles in NPC patients compared to controls. Receiver operating characteristic (ROC) analysis identified a combination of 8 metabolites as potential biomarkers for NPC. The machine learning analysis established a 8-metabolite NPC diagnostic model. The model demonstrated high accuracy (AUC=1) in distinguishing NPC patients from controls, which also performs well in the external validation set with an AUC value of 0.93, outperforming traditional methods that utilize plasma EBV-DNA copy number detection.
Conclusion: This study identified serum metabolites for NPC diagnosis, and the model developed provides a valuable tool for future clinical applications and further biological studies.