Rapid Identification and Quantification of Aqueous Antibiotics over a Machine Learning-Integrated Raman Sensor Array.

Area of Science:

• Analytical Chemistry
• Environmental Science
• Spectroscopy

Background:

• Single-substrate Surface-Enhanced Raman Spectroscopy (SERS) faces challenges in differentiating similar pollutants and achieving high accuracy.
• On-site detection of aqueous pollutants requires sensitive and selective analytical techniques.

Purpose of the Study:

• To develop a multisubstrate SERS fusion strategy integrated with machine learning (ML) for enhanced detection of sulfonamide antibiotics.
• To improve the accuracy and separability of analogous compounds in aqueous samples using SERS.

Main Methods:

• Fabrication of a SERS substrate array using gold nanoparticles modified with polyvinylpyrrolidone, l-cysteine, and 3-mercaptopropionic acid.
• Development of ML models using multi-interface concatenated spectra (MICS) for simultaneous identification and quantification.
• Validation of the method in real environmental water samples (river, lake, wastewater effluent).

Main Results:

• The multisubstrate SERS approach with ML achieved high classification accuracy (99.6% for single, 96.7% for mixtures) and quantitative prediction (R² up to 0.99).
• The method demonstrated robust performance in various water matrices with good recovery rates (median 97.5%).
• The entire process, from sample preparation to readout, takes approximately 30 minutes.

Conclusions:

• The proposed SERS fusion strategy coupled with ML offers a powerful paradigm for rapid, multianalyte detection of structurally similar pollutants.
• This approach significantly enhances SERS capabilities for on-site environmental monitoring, overcoming limitations of single-substrate methods.