Homogeneous multi-antibiotics residual identification in various actual water via SERS spectra multilayer perceptron...

Antibiotic residues pose varying degrees of potential hazards to the water environment and human health due to their diverse types. Surface-enhanced Raman spectroscopy (SERS) technology can achieve rapid detection of various antibiotic residues, and when combined with machine learning, it can identify multiple mixtures in Raman spectra. However, due to the interference of complex matrices and the similarity of the structure of the target itself, it is still a challenge for the detection and recognition of the same type of targets.

This study reported homogeneous multi-antibiotics SERS spectra recognition model for the identification of five sulfonamide antibiotics and their complex mixtures in various water by using a Multilayer Perceptron (MLP) algorithm. The results showed that the MLP model achieved accuracy, F1-score, 5-fold cross-validation (CV), and recall rates all above 90 %. Further, Gaussian Kernel Density Estimation (GKIM) was employed to augment the spectra of target substances containing ammonia nitrogen and phosphate and solved the reduction of model's identification accuracy when it used in nitrogen and phosphate matrices. The MLP model built on GKIM-augmented data achieved an antibiotic identification accuracy rate of 93 % in real water samples. The proposed model was validated in actual samples by LC-MS method and demonstrated excellent recognition capability.

This study demonstrates that combining MLP with GKIM can significantly enhance the model's ability to identify target substances in different water bodies, highlighting the broad application potential of this technology in environmental SERS detection and identification.