199: Explainable AI-Integrated SERS: A Framework for High-Accuracy Discrimination of E. Coli Pathotypes and Shigella Species

Introduction

Accurate and rapid discrimination of closely related bacterial pathogens is essential for effective surveillance, diagnostics, and treatment strategies. However, distinguishing pathogenic Escherichia coli and Shigella species has long posed significant challenges due to their extensive genetic and phenotypic similarities. In this study, we present an integrative analytical platform that combines Surface-Enhanced Raman Spectroscopy (SERS) with Artificial Intelligence (AI)-based classification to accurately differentiate five pathotypes of E. coli (enterotoxigenic, Shiga toxin–producing, enteropathogenic, enteroaggregative, and enteroinvasive) and four Shigella species.

Methods

A total of 334 strains were analyzed, yielding 7,679 SERS spectra subjected to rigorous preprocessing steps (baseline correction, de-spiking, peak binning, and standardization). We then evaluated four classification models, comprising two deep learning architectures (a One-Dimensional Convolutional Neural Network and a Multilayer Perceptron) and two traditional machine learning classifiers (Support Vector Machine and Random Forest). Additionally, feature importance analysis was performed using SHapley Additive exPlanations (SHAP) to interpret the contributions of key spectral features.

Results

The One-Dimensional Convolutional Neural Network emerged as the top performer with an overall classification accuracy of 97.7%, along with superior precision, recall, and F1 scores. We discovered that pivotal spectral features, analyzed using SHAP, were predominantly linked to protein and nucleic acid vibrations, offering insights into the subtle molecular differences among these closely related bacterial pathogens.

Significance

This integrated approach, leveraging SERS and explainable AI, demonstrates strong potential for high-throughput pathogen identification in microbiology, clinical diagnostics, and food safety management. Notably, given the absence of previous research capable of precisely differentiating E. coli pathotypes and Shigella spp., this approach holds significant promise for advancing both diagnostic practices and molecular-biological research.

Authors: Jun-Hyeok Ham, Hae-Yeong Kim