Core-shell MOF-SERS ratiometric sensing platform for volatile amine detection and machine learning-assisted meat freshness assessment

A novel ML-assisted SERS platform was developed for meat freshness evaluation. • Core-shell GNU/p-MBA@MIL-101 substrate enabled sensitive and stable SERS sensing. • Embedded p-MBA served as an internal standard and selectively reacted with VAs. • NH 3 , MA, PUT and CAD were quantified with ppb-level detection limits. • CNN analysis achieved 98.5% accuracy for meat freshness classification. Accurate and rapid evaluation of meat freshness remains challenging due to the complexity of spoilage markers and environmental interferences. Herein, a novel machine learning (ML)-assisted SERS platform was developed based on a core-shell GNU/p-MBA@MIL-101 substrate. Gold nanourchins (GNU) modified with p-mercaptobenzoic acid (p-MBA) were coated with MIL-101(Fe), enabling efficient gas capture and Raman signal enhancement. The embedded p-MBA was used as an internal standard (IS) and selectively reacted with volatile amines (VAs), to enable SERS detection with high sensitivity, reproducibility and anti-interference performance. Typical spoilage markers (NH 3 , MA, PUT and CAD) were quantitatively detected with limits of detection down to the ppb level. Convolutional neural networks (CNNs) were trained on SERS spectra collected from meat samples stored under different conditions, achieving a classification accuracy of 98.5%. This ML-assisted SERS strategy provides a reliable and accurate assessment of meat freshness, demonstrating strong potential for intelligent monitoring in food safety and quality control. Download: Download high-res image (119KB) Download: Download full-size image