Raman spectroscopy supported by machine learning reveals changes in balance of macromolecules in diabetic rat serum

In type 2 diabetes mellitus (T2DM), changes in glucose and lipid levels in serum are the first signals of disease progression. In this article, Raman spectroscopy supported with machine learning (ML) was used both to determine the level of imbalance between major metabolites caused by the disease and to investigate which metabolites were impacted the most. The principal component analysis separated the T2DM and control rats by PC1 and PC3. The bands that enabled that separation were assigned to all four main metabolite groups-amino acids (745-763 cm^-1, 985-1027 cm^-1), amides (1199-1258 cm^-1, 1574-1600 cm^-1, 1653-1712 cm^-1), polysaccharides (1104-1192 cm^-1), and lipids (1442-1472 cm^-1, 2831-3041 cm^-1). Moreover, diabetes induction caused major correlation changes between both proteins (amide) and the general amino acid band, with the organic matter band (CH band), which mainly arises due to lipid presence. Moreover, four different machine algorithms were employed to detect changes between the groups. AdaBoost showed the best performance in classifying the serum samples from diabetic and control rats (F₁ = 0.84), which was further refined when only the relevant bands were used for learning and classification (F₁ = 0.94). The bands responsible for the separation suggest that T2DM induces alterations in the protein profile, including changes in the levels of aromatic amino acids, as well as in both the quantity and composition of lipids, which is further confirmed by the ML classification.