Progress 07/01/23 to 06/30/24

Outputs
Target Audience:Target audience include students and faculty from academia, food scientists, professionals from the food industry, and regulators from government agencies. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?This project provided training for two doctoral students, one in food science, the other in computer science. How have the results been disseminated to communities of interest?News release (Feb. 26, 2024): Interdisciplinary research team uses AI to create revolutionary food safety technology (https://cafnr.missouri.edu/stories/interdisciplinary-research-team-uses-ai-to-create-revolutionary-food-safety-technology/) The results of this project have been disseminated in a top-tier peer-reviewed journal -Journal of Hazardous Materials (Impact factor = 13.6). It will also be disseminatedat the professional conferences including 2024 IFT conference, USDA PD meeting, and USDA Multistate NC-1194 annual meeting. What do you plan to do during the next reporting period to accomplish the goals?Next, we will focus on testing food samples contaminated with mixed pesticides by harnessing the enhanced sensitivity and specificity of SERS, coupled with the data-processing capabilities of machine learning algorithms. The collaboration between SERS and machine learning allows for the extraction of intricate spectral variances with unparalleled accuracy, facilitating the precise identification of pesticide compounds even in complex matrices.

Impacts
What was accomplished under these goals? Machine learning, a key subfield of artificial intelligence (AI), develops algorithms that automatically learn hidden patterns and relationships from datawithout explicit programming.During the past reporting year, weintroduced an innovative strategy for the rapid and accurate identification of pesticide residues in agricultural products by combining surface-enhanced Raman spectroscopy (SERS) with a state-of-the-art transformer model, termed SERSFormer.SERSFormerisbased on cutting-edge transformer technology that has achieved great success in large language models (e.g., ChatGPT) for natural language processing, providing a powerful synergy for accurate and efficient identification of pesticide residues. We developed SERSFormerconsisting of three main components: a task-specific embedding layer, a multi-tasking weight-sharing transformer encoder, and dedicated Multilayer Perceptron (MLP) heads preceding the output layers. Notably, SERSFormer encompasses two distinct branches for classification and regression, each equipped with an individual convolutional neural network (CNN) embedder of a similar design. The Transformer Encoder comprises multiple Multi-head Attention Encoder layers strategically shared by both the regression and classification branches. The classification branch extends into a two-layered MLP head, predicting outcomes across six distinct classes. Gold-silver core-shell nanoparticles were synthesized and served as high-performance SERS substrates, which possess well-defined structures, uniform dispersion, and a core-shell composition with an average diameter of 21.44 ± 4.02 nm, as characterized by TEM-EDS. The SERSFormer model demonstrated exceptional proficiency in qualitative analysis, successfully classifying six categories, including five pesticides (coumaphos, oxamyl, carbophenothion, thiabendazole, and phosmet) and a control group of spinach data, with 98.4% accuracy. For quantitative analysis, the model accurately predicted pesticide concentrations (mean absolute error of 0.966, mean squared error = 1.826, and R2score = 0.849). This study synergizes cutting-edge machine learning models and advanced SERS techniques to rapidly and accurately detect pesticide residues in agricultural products. Integrating SERS and the SERSFormer model demonstrates the potential to transform pesticide analysis with high sensitivity, specificity, and efficiency. The SERSFormer serves as a versatile tool for both qualitative and quantitative analysis. Qualitatively, it accurately identified six pesticide categories, benefiting significantly from preprocessing techniques like noise reduction, baseline correction, and normalization. Quantitatively, the model excelled in predicting pesticide concentrations. The study underscores the significance of data normalization techniques in pesticide classification and quantification tasks, including log-min-max normalization, log-convolution, and confusion matrix analysis. This integrated approach, combining SERS with machine learning, offers a promising route for rapid, reliable pesticide detection, with significant implications for monitoring food safety in the agriculture and food sectors.

Publications

• Type: Journal Articles Status: Published Year Published: 2024 Citation: Hajikhani, M., Hegde, A., Snyder, J., Cheng, J., Lin, M. 2024. Integrating transformer-based machine learning with SERS technology for the analysis of hazardous pesticides in spinach. Journal of Hazardous Materials. 134208. (https://doi.org/10.1016/j.jhazmat.2024.134208) (Impact factor = 13.6)