Characterization of Serum Biochemical Changes in Response to Alcohol Consumption Using FT-Raman Spectroscopy

Alcohol is one of the most widely used stimulants globally, with increasing consumption rates annually. Prolonged alcohol consumption leads to significant physiological and neurological changes, potentially resulting in various diseases, which could be detect in serum. Therefore, FT-Raman spectroscopy was used to measure chemical changes in the serum of heavy drinkers, and these results were compared with the serum composition of a control group of non-drinkers. Obtained results showed a significant increase in lipids and a decrease in the amide/lipid ratio after alcohol administration. Additionally, higher amide III/amide II and lower amide II/amide I ratios were observed in the serum of alcohol-addicted patients. These shifts of 892, 966, 1286, 1459 and 2940 cm-1 peaks indicate alterations in the protein-lipid balance due to alcohol consumption. Principal component analysis (PCA) was employed to further analyze the spectral data, with the first three principal components accounting for 95.36% of total data variability. A Bayesian-optimized k-nearest neighbor (BO-KNN) model was applied for classification. The optimal hyperparameters-two neighbors, correlation distance metric, and squared inverse distance weight-were determined after seven iterations, resulting in a remarkably low classification error of 0.0384. The BO-KNN model achieved training and test accuracies of 95.92% and 90.48%, respectively. Receiver operating characteristic (ROC) analysis showed an area under the curve (AUC) of 1.00 for the training set and 0.90 for the test set, demonstrating the model's high precision and robustness. In conclusion, FT-Raman spectroscopy effectively revealed significant chemical changes in the serum of alcohol-addicted individuals, providing valuable insights into the biochemical impact of alcohol consumption. The study used FT-Raman spectroscopy to detect chemical changes in the serum of heavy drinkers, revealing an increase in lipids and shifts in protein-lipid balance versus control. Peaks at 892, 966, 1286, 1459, and 2940 cm-1 indicated these alterations. Principal component analysis (PCA) explained 95.36% of data variability. A Bayesian-optimized k-nearest neighbor (BO-KNN) model classified the data with high accuracy (95.92% training and 90.48% test) and a low error rate (0.0384). ROC analysis demonstrated excellent model performance, with an AUC of 1.00 for the training set and 0.90 for the test set.image