Untargeted metabolomics and machine learning reveal novel non-invasive biomarker panels for diagnosis and risk stratification in extranodal NK/T-cell lymphoma

Abstract Extranodal NK/T-cell lymphoma (NKTCL) is a rare and highly aggressive malignancy with distinct epidemiological features, posing significant challenges for early accurate diagnosis and prognosis assessment. There is an urgent need to develop effective non-invasive biomarkers and precise therapeutic strategies. Currently, research on circulating metabolic biomarkers for early diagnosis and patient risk stratification remains insufficient. In this study, we conducteduntargeted metabolomic analysis on plasma samples from a total of 821 individuals consisting of 247 NKTCL patients and 574 normal controls, systematically characterizing the metabolic reprogramming features of NKTCL. Utilizing machine learning algorithms, we constructed a diagnostic model for NKTCL. This model demonstrated excellent performance in training and validation cohort（Figure1）. Furthermore, the prognostic model for NKTCL developed using machine learning also exhibited superior predictive accuracy（Figure2）. It effectively stratified patients into distinct risk subgroups with significantly different survival outcomes, thereby providing a basis for precision interventions. Collectively, this study delineates the metabolic landscape of NKTCL and identifies two novel metabolite biomarker panels enabling non-invasive early diagnosis and prognostic risk stratification, respectively, laying the foundation for advancing precision medicine in NKTCL.