Metabolic processes monitored by MRS precede micro-structural changes visualised by imaging. The high noise and the overlapping spectra of metabolites affect the accurate quantification of metabolite’s concentration. This work hypothesizes that each tissue has a unique metabolic fingerprint and a diagnostic model could be built based on tissue spectra. The MRS datasets however are usually small and acquired with different parameters. This work quantum-mechanically simulates spectra and uses the augmented spectra to train a model that can differentiate metastasis from glioblastoma brain cancer. The trained model was tested on acquired spectra from the INTERPRET single-voxel dataset and illustrated a ROC-AUC=0.90.