S.aureus-SERS-AI

Project Overview

This repository contains the Surface-Enhanced Raman Spectroscopy (SERS) dataset and machine learning code used to classify Staphylococcus aureus isolates into MRSA, ERSA, and SSA. The dataset accompanies the manuscript:

“Standardization is the Key to Reliable SERS-Based Machine Learning for Antimicrobial Resistance Classification.

 What This Project Does

Provides a SERS dataset consisting of three CSV files (MRSA, ERSA, SSA), each containing five isolates as separate sheets.

Includes preprocessing, feature selection, and machine learning scripts.

Demonstrates the effect of 12 data partitioning strategies on model performance.

Enables reproducible classification of AMR-related S. aureus isolates.

 Why This Project Is Useful

Offers one of the few public SERS datasets with clearly defined clinical isolates.

Helps researchers evaluate standardization issues in SERS-based machine learning.

Provides ready-to-run ML pipelines for:

Preprocessing spectra

Applying Boruta, LASSO, and PCA

Training RF, SVM, kNN, NNET, and XGB models

Serves as a foundation for benchmarking, clinical AMR studies, or SERS–AI algorithm development.

 Dataset Description

The dataset consists of ~15,000 SERS spectra collected from:

5 MRSA isolates

5 ERSA isolates

5 SSA isolates

Each isolate contributes ~1,000 spectra.

Spectral Range

400–1800 cm⁻¹

818 Raman shift positions

785 nm excitation wavelength

1-second integration time

File Format

Each .csv file contains:

Rows: individual spectra

Columns: Raman intensities for each wavenumber