An approach in developing graphical feature maps derived from machine learning and its application in loquat juice...

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Machine Learning

Background:

• Accurate classification of fruit juice varieties is crucial for quality control and authenticity verification.
• Traditional methods often lack the precision to differentiate closely related cultivars based on complex chemical profiles.
• Developing advanced analytical techniques is essential for nuanced food analysis.

Purpose of the Study:

• To introduce a novel methodology for generating graphical feature maps using weighted artificial neural networks (w-ANNs).
• To apply this methodology for the classification of loquat juice varieties (loquat_baisha and loquat_hongsha) using a convolutional neural network (CNN).
• To identify key chemical compounds and molecular feature descriptors that differentiate loquat varieties.

Main Methods:

• Weighted artificial neural networks (w-ANNs) for graphical feature map generation.
• Convolutional neural network (CNN) with TensorFlow (TF) for classification.
• Headspace gas chromatograph-ionic mass spectroscopy (HS-GC-IMS) for chemical compound identification.
• SHapley Additive exPlanations (SHAP) for identifying influential molecular feature descriptors (MFDs).
• Development of a loquat chemical library using PubChem data.

Main Results:

• HS-GC-IMS identified distinct sets of seven key compounds for loquat_baisha and loquat_hongsha.
• SHAP analysis highlighted Kappa2, Gasteiger charge, and LogP as significant MFDs for loquat_baisha.
• Kappa2, Kappa3, and Fraction_SP3 were identified as key MFDs for loquat_hongsha.
• Graphical feature maps were successfully constructed to aid classification.
• A comprehensive loquat chemical library was established.

Conclusions:

• The developed methodology provides a novel approach for classifying loquat juice varieties based on chemical profiles.
• The identified key compounds and MFDs offer insights into the chemical distinctions between loquat_baisha and loquat_hongsha.
• The effectiveness of the methodology is context-dependent, requiring evaluation for different applications.