Investigation of biochemical alterations in the brain of db/db diabetic mice using integrated FTIR and Raman spectroscopy combined with machine learning.

This study systematically analyzed the molecular metabolic alterations in the brain tissue of type 2 diabetic mice at different disease stages by integrating Fourier transform infrared (FTIR) and Raman spectroscopy with machine learning approaches. The results demonstrated glycogen/polysaccharide accumulation, stage-specific remodeling of amide bands, early instability of β-sheet structures, progressive collagen deposition, and lipid metabolic disturbances, with a transient compensatory phase observed prior to 12 weeks. FTIR was particularly sensitive to polar groups and total lipid content, whereas Raman spectroscopy provided insights into protein conformations and lipid unsaturation. Together, the two modalities revealed the molecular remodeling features of diabetic brain tissue. Partial least squares-discriminant analysis (PLS-DA) based on fused spectra achieved high classification accuracy across disease stages (overall accuracy: 88.9%), and variable importance projection (VIP) analysis identified 1180-1203 cm-1 (FTIR) and 1300 and 2853 cm-1 (Raman) as critical discriminatory bands. Furthermore, a GA-PLS regression model, retaining only 7% of the spectral variables, effectively predicted the severity of diabetic brain injury (R2 = 0.885, RMSE = 0.407), revealing a linear metabolic trajectory across disease progression. Collectively, this spectroscopic-computational strategy enables sensitive detection of stage-dependent metabolic alterations in diabetic encephalopathy and provides a novel framework for identifying characteristic optical biomarkers and elucidating potential molecular mechanisms, while also offering new insights for precise cause-of-death determination in forensic medicine.