Rapid identification of Litopenaeus vannamei pathogenic bacteria: a combined approach using surface-enhanced Raman spectroscopy (SERS) and deep learning.

Abstract

Pathogenic bacterial infections are one of the leading causes of mortality in Litopenaeus vannamei, seriously affecting the economic efficiency of the shrimp aquaculture industry. However, traditional pathogen detection methods, such as the polymerase chain reaction (PCR), have drawbacks, including complex procedures and long processing times. Raman spectroscopy provides valuable biomolecular feature information and, when combined with deep learning, enables the detection of pathogens. However, the limited availability of spectral data hinders model performance. Therefore, we proposed a novel method that integrated surface-enhanced Raman spectroscopy (SERS), least-squares generative adversarial network (LSGAN), and Transformer to achieve high-precision identification of four common shrimp pathogens. This method employed LSGAN to generate synthetic spectra resembling real spectra for data augmentation and utilized the Transformer for high-precision identification of pathogens. First, the original dataset of 160 spectra was expanded to 2160 using LSGAN. It was shown that the LSGAN-enhanced data could effectively improve the classification performance of the Transformer, and the accuracy of the Transformer in the classification task of shrimp pathogens was 99.69%, which was 2.82% higher than that before the data enhancement. Additionally, Transformer achieved a classification accuracy of 91.04% on a publicly available microbial Raman spectral dataset, demonstrating strong generalization capability. Our research introduces novel insights into the classification of limited Raman spectra and presents a rapid, accurate method for detecting pathogens in shrimp farming, aiding early disease prevention and control.