Machine Learning-Enhanced Cerebrospinal Fluid N-Glycome for the Diagnosis and Prognosis of Primary Central Nervous System Lymphoma

The diagnosis and prognosis of Primary Central Nervous System Lymphoma (PCNSL) present significant challenges. In this study, the potential use of machine learning algorithms in diagnosing and predicting the prognosis for PCNSL based on cerebrospinal fluid (CSF) N-glycomics was investigated. First, CSF samples obtained from a cohort of 60 PCNSL patients and 30 controls were analyzed by hydrophilic interaction-based ultra performance liquid chromatography (HILIC-UPLC)-fluorescence mass spectrometry. Subsequently, nine machine learning models were established to diagnose PCNSL based on the changes of CSF N-glycome, with the Random Forest algorithm proving to be the most effective, achieving an accuracy of 100% in the training set and 89.3% in the test set. Moreover, a COX proportional-hazard model and a nomogram incorporating CSF N-glycome (GP6 and GP27) along with clinical data (age) were crafted. This nomogram's discrimination capacity was considered satisfactory, as evidenced by a C-index of 0.804 (95% CI: 0.68, 0.927). The study reveals that machine learning models based on CSF N-glycome offer a valuable approach for diagnosing and prognosticating PCNSL, demonstrating high accuracy and sensitivity in both classification and survival analysis. These findings may offer new insights into the molecular mechanisms underlying PCNSL and contribute to the advancement of personalized medicine for patients with this disease.