Structural Analysis and Classification of Low-Molecular-Weight Hyaluronic Acid by Near-Infrared Spectroscopy: A...

Area of Science:

• Biochemistry and Spectroscopy
• Machine Learning Applications in Material Science

Background:

• Low-molecular-weight hyaluronic acid (LMWHA) derived from acid degradation (LMWHA-A) and enzymatic hydrolysis (LMWHA-E) possess distinct structures.
• Misidentification of LMWHA sources poses significant health hazards and commercial risks due to differing properties and applications.

Purpose of the Study:

• To elucidate the structural disparities between LMWHA-A and LMWHA-E.
• To develop a rapid and accurate classification method for distinguishing LMWHA-A from LMWHA-E using near-infrared (NIR) spectroscopy and machine learning.

Main Methods:

• Structural analysis employed Nuclear Magnetic Resonance (NMR), Fourier Transform Infrared (FTIR) spectroscopy, and aquaphotomics.
• Near-infrared (NIR) spectroscopy combined with 2D correlation analysis (2DCOS) was utilized for spectral data acquisition.
• Dimensionality reduction (PCA, KPCA, t-SNE) and classification algorithms (PLS-DA, SVC, RF, 1D-CNN, LSTM) were evaluated.

Main Results:

• Deep learning models, specifically one-dimensional convolutional neural network (1D-CNN) and long short-term memory (LSTM), achieved 100% accuracy in classifying LMWHA types on both training and test datasets.
• Traditional machine learning methods, including Genetic Algorithm-Support Vector Classification (GA-SVC) and Random Forest (RF), showed a maximum accuracy of 90% on the test dataset.
• Deep learning approaches demonstrated superior performance compared to traditional machine learning for LMWHA classification.

Conclusions:

• Deep learning models offer a highly accurate and efficient method for differentiating between acid-degraded and enzymatically hydrolyzed LMWHA.
• This research establishes a novel methodological framework for the rapid and precise classification of biological macromolecules like LMWHA.