Improving the detection accuracy of the dual SERS aptasensor system with uncontrollable SERS “hot spot” using machine learning tools

Simultaneous detection of food contaminants is crucial in addressing the collective health hazards arising from the presence of multiple contaminants. However, traditional multi-competitive surface-enhanced Raman scattering (SERS) aptasensors face difficulties in achieving simultaneous accurate detection of multiple target substances due to the uncontrollable SERS "hot spot". In this study, using chloramphenicol (CAP) and estradiol (E2) as two target substances, we introduced a novel approach that combines machine learning methods with a dual SERS aptasensor, enabling simultaneous high-sensitivity and accurate detection of both target substances. The strategy effectively minimizes the interference from characteristic Raman peaks commonly encountered in traditional multi-competitive SERS aptasensors. For this sensing system, the Au@4-MBA@Ag nanoparticles modified with sulfhydryl (SH)-CAP aptamer and Au@DTNB@Ag NPs modified with sulfhydryl (SH)-E2 aptamer were used as signal probes. Additionally, Fe3O4@Au nanoflowers integrated with SH-CAP aptamer complementary DNA and SH-E2 aptamer complementary DNA were used as capture probes, respectively. When compared to linear regression random forest, and support vector regression (SVR) models, the proposed artificial neural network (ANN) model exhibited superior precision, demonstrating R2 values of 0.963, 0.976, 0.991, and 0.970 for the training set, test set, validation set, and entire dataset, respectively. Validation with ten spectral groups reported an average error of 244 μg L-1. The essence of our study lies in its capacity to address a persistent challenge encountered by traditional multiple competitive SERS aptasensors – the interference generated by uncontrollable SERS "hot spot" that hinders simultaneous quantification. The accuracy of the predictive model for simultaneous detection of two target substances was significantly improved using machine learning tools. This innovative technique offers promising avenues for the accurate and high-sensitive simultaneous detection of multiple food and environment contaminants.