Aggregation- enhanced SERS of 3-MPA/4-MPY modified au concave Nanocubes coupled with CNN-PLS algorithm for quantitative detection of soluble Hg²⁺ in cinnabar

Cinnabar, with low water solubility and poor oral absorption, poses challenges in toxicity assessment. Relying solely on total mercury leads to overestimation, making detection of soluble Hg²⁺ essential. This study presents a surface-enhanced Raman scattering (SERS) strategy using Au concave nanocubes (Au CNC) modified with 3-mercaptopropionic acid (3-MPA) and 4-mercaptopyridine (4-MPY), combined with a data-augmented convolutional neural network-partial least squares regression (CNN-PLS) model, for quantifying soluble Hg²⁺ in cinnabar and its preparations. 3-MPA binds Hg²⁺, inducing Au CNC aggregation and enhances the SERS signal, while 4-MPY improves sensitivity. To overcome insufficient data in practical, data augmentation was used to expand the SERS dataset, and the performance of CNN, PLS, and CNN-PLS models was compared before and after augmentation. The results indicate that the data-augmented CNN-PLS model performed best, with a coefficient of determination (R²) of 0.999 in the range of 0.001–1 μg/mL and a Mean Squared Error of Prediction (MSEP) of 0.0001. This method was applied to quantify Hg²⁺ leaching from cinnabar and four cinnabar-containing preparations after 12 h of immersion in artificial gastric juice. The mean Hg²⁺ concentrations from three cinnabar batches were 0.691, 0.557, and 0.546 μg/mL, while for the preparations, the concentrations were 0.085, 0.047, 0.038, and 0.023 μg/mL. These results showed excellent agreement with atomic fluorescence spectrometry (AFS), confirming the method's reliability and accuracy. This simple, accurate, and safe approach offers a novel solution for mercury safety assessment in cinnabar and its preparations, with significant potential for traditional Chinese medicine safety evaluation and environmental monitoring.