Antibiotic SERS spectral analysis based on data augmentation and attention mechanism strategy

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Machine Learning

Background:

• Raman spectrum data analysis is increasingly using machine learning.
• Challenges include acquiring large datasets and preserving spectral information.

Purpose of the Study:

• To develop a machine learning strategy for analyzing antibiotic spectral data.
• To address data scarcity and information loss in spectral analysis.

Main Methods:

• Utilized a Generative Adversarial Network (GAN) to amplify Surface-Enhanced Raman Spectroscopy (SERS) data of eight antibiotics tenfold.
• Implemented a one-dimensional convolutional neural network (1D-CNN) with an attention mechanism module for enhanced feature extraction.

Main Results:

• The 1D-CNN achieved 97.5% accuracy in classifying eight individual antibiotics.
• An accuracy of 89.4% was obtained for classifying four mixtures within the same antibiotic class.

Conclusions:

• The combined approach of data amplification and attention mechanisms effectively improves the analysis of antibiotic spectral data.
• This strategy enhances the capability of machine learning models in spectral classification tasks.