DeepRaman: Implementing surface-enhanced Raman scattering together with cutting-edge machine learning for the...

Area of Science:

• Biotechnology and Biomedical Engineering
• Spectroscopy and Analytical Chemistry

Background:

• Bacterial endotoxins, lipopolysaccharides from Gram-negative bacteria, are key biomarkers for identification.
• Surface-Enhanced Raman Scattering (SERS) with silver nanorod substrates was used to obtain spectral data.
• Datasets included eleven endotoxins and three control substances (chitin, LTA, PGN).

Purpose of the Study:

• To develop and evaluate DeepRaman, a novel architecture for classifying raw SERS Raman spectra from biological materials.
• To compare DeepRaman's performance against traditional machine learning algorithms and CNNs.
• To assess DeepRaman's ability to function independently and with smaller datasets.

Main Methods:

• Utilized DeepRaman, inspired by Progressive Fourier Transform and scalogram transformation.
• Employed SERS spectroscopy for spectral data acquisition.
• Compared DeepRaman with classical machine learning (SVM, k-NN, RF) and modified deep learning approaches.

Main Results:

• DeepRaman achieved an exceptional accuracy of 100% in classifying bacterial endotoxins.
• Traditional machine learning algorithms also showed high accuracies, exceeding 99%.
• DeepRaman demonstrated superior performance, especially with smaller datasets and challenging spectral baselines.

Conclusions:

• DeepRaman offers precise endotoxin classification with high accuracy, outperforming conventional methods.
• The architecture's autonomous operation and efficiency with small datasets facilitate broader clinical adoption.
• DeepRaman holds significant potential for accelerating medical diagnostics and treatment decisions in pathogenic infections.