Handling multiblock data in wine authenticity by sequentially orthogonalized one class partial least squares.

Area of Science:

• Chemometrics
• Food Science
• Analytical Chemistry

Background:

• Food authentication is crucial for quality control and consumer protection.
• Integrating multi-source data can improve the accuracy of analytical models.
• Existing methods may struggle to effectively utilize information from diverse datasets.

Purpose of the Study:

• To propose and evaluate a novel data fusion method for food authentication.
• To assess the effectiveness of OC-PLS in handling multi-source data.
• To differentiate high-quality Slovak Tokaj wines from lower-value wines using spectroscopic data.

Main Methods:

• Developed a new approach called Orthogonal projection to Latent Structures (OC-PLS).
• Incorporated an orthogonalization step to remove redundant information between datasets.
• Defined an acceptance area for a target class using OC-PLS.
• Applied the method to simulated data and real-world spectroscopic (UV-VIS, IR) data.

Main Results:

• OC-PLS demonstrated superior performance compared to using individual data blocks in both simulated and real-world case studies.
• The method effectively identified common and distinct information across different data sources.
• Achieved an optimal balance between sensitivity and selectivity in the prediction phase.

Conclusions:

• The proposed OC-PLS method is effective for food authentication by integrating multi-source data.
• This approach offers improved accuracy and better information utilization compared to traditional methods.
• OC-PLS provides a robust framework for quality assessment in food products.