Multi-omics and machine learning approaches uncover metabolic biomarkers for diagnosis of germinoma

Metabolomics and a multi-step machine-learning approach were performed to identify potential diagnostic metabolic biomarkers in the cerebrospinal fluid of 65 patients (26 germinoma, 31 non-iGCT, 8 NGGCT). Among them, six germinoma, nine non-iGCT, four NGGCT plus the held-out test set from nested 5 fold cross-validation (random eight samples) served as independent test data to evaluate classification models. A multi-omics strategy was employed to further investigate the pathogenesis of germinoma using tissue samples, including 17 transcriptome samples (3 germinoma and 3 NGGCT as validation), 7 metabolomic samples and 58 DNA methylation samples, focusing on dysregulated metabolic pathways and immune cell infiltration. 51 differential metabolites were detected between germinoma and non-iGCT.