Explainable Deep Learning to Predict Kelp Geographical Origin from Volatile Organic Compound Analysis.

Area of Science:

• Food Science
• Analytical Chemistry
• Computational Biology

Background:

• Consumer choices increasingly depend on kelp origin, beyond flavor and nutrition.
• Limited research exists on kelp origin traceability using volatile organic compounds (VOCs).
• The application of deep learning for traceability is hindered by its 'black-box' nature.

Purpose of the Study:

• To develop a method for identifying kelp origin using VOC analysis.
• To apply explainable deep learning for transparent and accurate traceability.
• To enhance consumer trust and practical application of kelp origin information.

Main Methods:

• Identified 115 distinct VOCs in kelp samples using gas chromatography-ion mobility spectroscopy (GC-IMS).
• Developed a one-dimensional convolutional neural network (1D-CNN) model incorporating 107 key VOCs.
• Utilized SHapley Additive exPlanations (SHAP) for model interpretability.

Main Results:

• Achieved perfect performance metrics: 100% accuracy, precision, recall, F1 score, and 1.0 AUC.
• Successfully discerned kelp origin based on VOC profiles.
• Identified specific VOCs like 1-Octen-3-ol-M and (+)-limonene as key indicators.

Conclusions:

• Explainable deep learning effectively traces kelp origin via VOC analysis.
• The model provides accurate and interpretable results for geographic traceability.
• This approach can increase consumer confidence and facilitate real-world kelp origin verification.