Raman Spectroscopy Meets Machine Learning: From Data Generation to Molecular Interpretation

Abstract

Raman spectroscopy provides rich molecular information about complex biological and polymeric systems, yet subtle spectral changes, especially those induced by low-concentration additives or enzymatic modifications, often remain undetectable by conventional analysis. This work demonstrates how machine learning, particularly tree-based ensemble models, can address both the challenge of limited training data and the need for interpretable classification.

We propose a PCA-based generative approach that synthesizes realistic Raman spectra by sampling principal component scores from normal distributions, enabling dataset augmentation even when experimental measurements are scarce. Applied to adipose tissue before and after lipase treatment, this method achieved 90.6% classification accuracy on synthetic data, confirming its utility for expanding spectral libraries.

For classification and interpretation, we systematically compared Random Forest, Gradient Boosting, AdaBoost, Voting, and Stacking ensembles. On whey protein isolate–hyaluronic acid conjugates (HA concentrations 0–0.5 wt%), the Stacking classifier achieved 82.0% balanced accuracy with 95.4% sensitivity and 68.7% specificity, while Stacking regression yielded R² = 0.764 for HA quantification. Feature importance analysis revealed that phenylalanine (1003 cm⁻¹), NH₃⁺ rocking (1125 cm⁻¹), C–C stretching (1206 cm⁻¹), and amide III (1240 cm⁻¹) were most influential for differentiation. In lipase-treated adipose tissue, Gradient Boosting identified the C=C stretching band at 1650 cm⁻¹ as the dominant spectral marker (95.4% importance), consistent with triglyceride hydrolysis to free fatty acids and mono/diglycerides.

These findings establish ensemble learning as a powerful, interpretable intermediate between classical chemometrics and deep learning, with direct applicability to quality control in food and pharmaceutical industries.

This research was funded by the Russian Ministry of Science and Higher Education (state assign-ment no. FSMG-2025-0054)

Speaker

Ekaterina S. Prikhozhdenko
Laboratory of medical equipment in the field of in vitro diagnostics, Moscow Institute of Physics and Technology, Dolgoprudny, Moscow region, Russia
Russia

Discussion

Ask question