Explainable AI-SERS approach for highly accurate discrimination of Escherichia coli pathotypes and Shigella species

Pathogenic Escherichia coli and Shigella species cause severe diarrheal diseases with high mortality but remain difficult to distinguish using conventional methods due to their close genetic and proteomic relatedness. To address this challenge, we propose an explainable artificial intelligence (XAI) with surface-enhanced Raman spectroscopy (SERS) platform for rapid and accurate identification of E. coli pathotypes and Shigella species. We generated 7819 SERS spectra from 294 strains, including 195 representing five pathotypes of E. coli and 99 of Shigella species. This dataset was analyzed within an XAI framework using deep learning models, including a one-dimensional convolutional neural network (1D-CNN) and a multilayer perceptron, and compared with traditional machine learning classifiers. The 1D-CNN achieved 97.7% accuracy, outperforming conventional classifiers. SHapley Additive exPlanations analysis revealed the specific features and molecular components contributing to classification, providing biochemical interpretability. This study demonstrates the potential of XAI-SERS for precise, explainable identification of E. coli pathotypes and Shigella species.