Glycogen structure is closely associated with its physiological functions. Previous studies confirmed that liver glycogen structure had two dominant states: mainly stable during the day and largely fragile at night. However, the diurnal change of glycogen structure is impaired, with dominant fragility in diseased conditions such as diabetes mellitus and liver fibrosis. Therefore, the persistent structural fragility of glycogen particles could be a potential molecular-level pathological biomarker for early screening of certain liver diseases. However, the current method for identifying glycogen structural stability and fragility suffers from sophisticated procedures and reliance on expensive instruments, which demands developing novel methods for rapidly discriminating the two types of glycogen particles. This study applied surface-enhanced Raman spectroscopy (SERS) to generate SERS spectra of glycogen samples, revealing distinct structural differences between fragile and stable glycogen particles. Machine learning models were then constructed to predict the structural states of unknown glycogen samples via SERS spectra, according to which the convolutional neural network (CNN) model achieved the best discrimination capacity. Taken together, the SERS technique coupled with the CNN model can identify stable and fragile liver glycogen samples, facilitating the application of glycogen structural fragility as a biomarker in diagnosing liver diseases.
Keywords: Deep learning; Diurnal change; Glycogen; Raman spectrometry; Raman spectroscopy; Structural fragility.
Copyright © 2024 Elsevier B.V. All rights reserved.