In this study, a non-labeled sensor system for direct determining human glycated albumin levels for medical application is proposed. Using machine learning methods applied to surface-enhanced Raman scattering (SERS) spectra of human glycated albumin and serum human albumin enabled the avoidance of complex sample preparation. By implementing linear discriminant analysis and regularized linear regression, classification and regression problems were solved based on the spectra obtained as a result of the experiment. The results show that, coupled with data augmentation and a special cross-validation procedure, the methods we employed yield better results in the corresponding tasks in comparison with popular random forest methods and the support vector method. The results show that SERS, in combination with machine learning methods, can be a powerful and effective tool for the simple and direct assay of protein mixtures.

Angiotensin I-converting enzyme (ACE) is a peptidase widely presented in human tissues and biological fluids. ACE is a glycoprotein containing 17 potential N-glycosylation sites which can be glycosylated in different ways due to post-translational modification of the protein in different cells. For the first time, surface-enhanced Raman scattering (SERS) spectra of human ACE from lungs, mainly produced by endothelial cells, ACE from heart, produced by endothelial heart cells and miofibroblasts, and ACE from seminal fluid, produced by epithelial cells, have been compared with full assignment. The ability to separate ACEs’ SERS spectra was demonstrated using the linear discriminant analysis (LDA) method with high accuracy. The intervals in the spectra with maximum contributions of the spectral features were determined and their contribution to the spectrum of each separate ACE was evaluated. Near 25 spectral features forming three intervals were enough for successful separation of the spectra of different ACEs. However, more spectral information could be obtained from analysis of 50 spectral features. Band assignment showed that several features did not correlate with band assignments to amino acids or peptides, which indicated the carbohydrate contribution to the final spectra. Analysis of SERS spectra could be beneficial for the detection of tissue-specific ACEs.

Specific monosaccharide residue, β-D-galactofuranose (Galf) featuring a five-membered ring structure, is found in the glycans of fungi and bacteria, but is normally absent in healthy mammals and humans. In this study, synthetic oligosaccharides mimicking bacterial and fungal glycans were investigated by SERS (Surface-Enhanced Raman Scattering) techniques for the first time to distinguish between different types of glycan chains. SERS spectra of oligosaccharides related to fungal α-(1→2)-mannan, β-(1→3)-glucan, β-(1→6)-glucan, galactomannan of Aspergillus, galactan I of Klebsiella pneumoniae, and diheteroglycan of Enterococcus faecalis were measured. To analyze the spectra, a number of machine learning methods were used that complemented each other: principal component analysis (PCA), confidence interval estimation (CIE), and logistic regression with L1 regularization. Each of the methods has shown own effectiveness in analyzing spectra. Namely, PCA allows the visualization of the divergence of spectra in the principal component space, CIE visualizes the degree of overlap of spectra through confidence interval analysis, and logistic regression allows researchers to build a model for determining the belonging of the analyte to a given class of carbohydrate structures. Additionally, the methods complement each other, allowing the determination of important features representing the main differences in the spectra containing and not containing Galf residue. The developed mathematical models enabled the reliable identification of Galf residues within glycan compositions. Given the high sensitivity of SERS, this spectroscopic technique serves as a promising basis for developing diagnostic test systems aimed at detecting biomarkers of fungal and bacterial infections.

In this study, we used the SERS method for the first time to measure the spectra of four polysaccharide markers of fungal infections: linear β-(1→3)- and β-(1→6)-linked D-glucans, branched mannan of Candida albicans and galactomannan of Aspergillus fumigatus. Aqueous solutions of the polysaccharides were studied in concentrations from 10 pg/mL to 100 μg/mL. The spectra were analyzed using machine learning methods: principal component analysis for data visualization and partial least squares with a ridge regularizer, which were used to construct metrics reflecting the accuracy of substance recognition relative to each other. The spectral changes with varying analyte concentration were observed and stable calibration has been achieved. Subsequent measurements of fungal polysaccharides in the presence of a physiological concentration of human serum albumin (45 mg/mL), used to model blood serum, enabled accurate analyte detection in a clinically relevant concentration range of 10 pg/mL to 100 ng/mL. In this case, the calibration dependence was calculated using the partial least squares method with the L1-regularizer. Blind testing was evaluated using a train-derived applicability-domain criterion based on the disagreement between the model prediction and an independent concentration estimate.