A statistical framework for biomarker identification of biopsies using HR-MAS HSQC spectroscopy

Cancer is one of the principal causes of morbidity and mortality worldwide. One of the strategies employed by the emergent science of metabolomics is cancer biomarker extraction. In this context, the technique of High-Resolution Magic Angle Spinning (HR-MAS) Nuclear Magnetic Resonance (NMR) spectra is widely used in metabolomic analysis involving tissue studies. Indeed, the NMR offers the potential to study molecular structures and their associations and interactions. In this paper, we develop a novel scheme for biomarker identification from 2D NMR spectrum. The biomarker identification is obtained by comparing 2D NMR spectral patterns in the NMR spectrum of the biopsy with specific library coding reference spectra of pure metabolites. Our comparison model is improved by combining probability and fuzzy theories to represent uncertainty and fuzzyness with our inference model. Validation experiments show that the proposed algorithm provides more accurate metabolite identification than the classical Support Vector Machine (SVM) method.