Researchers have developed a real-time one-class classification (OCC) method that can detect adulterated edible oils—like avocado oil—without relying on pre-built models or prior knowledge of contaminants.
Study: catGRANULE 2.0: accurate predictions of liquid-liquid phase separating proteins at single amino acid resolution. Image Credit: New Africa/Shutterstock.com
In a study published in Foods, the team demonstrated how analyzing residuals from thousands of Monte Carlo-sampled models comparing authentic and suspect samples allowed them to successfully identify six out of 40 adulterated avocado oil products. These findings were later confirmed through chemical marker analysis.
Background
Food fraud, particularly economically motivated adulteration (EMA) of high-value oils such as avocado oil, remains a pressing issue for both consumer health and industry integrity. Traditional detection methods—including spectroscopy and chromatography-mass spectrometry—often rely on two-class classification models that require known adulterants, limiting their effectiveness in real-world scenarios where the contaminant is unknown.
Recent advances using one-class classification (OCC) models, such as data-driven soft independent modeling of class analogy (DD-SIMCA) and one-class partial least squares (OCPLS), have improved authenticity testing. However, these models still depend on access to globally representative sample sets, which are difficult to collect, validate, and maintain. This issue becomes especially critical during blind inspections, where the specific adulterant is not known in advance.
Avocado oil, known for its nutritional benefits and premium price point, is often a target for adulteration with cheaper oils like soybean, corn, or rapeseed. Existing methods continue to face challenges when dealing with unknown adulterants and unrepresentative training data.
This study proposes a new solution: a real-time OCC modeling approach using Monte Carlo sampling and absolute centered residual (ACR) analysis. By probabilistically identifying “good” models built exclusively from authentic samples, the method flags adulterated products without relying on pre-trained models or predefined adulterants. The results were validated using chemical markers, offering a more adaptable and reliable strategy for high-value oil authentication.
A Closer Look at the Method
To test the approach, researchers analyzed 40 avocado oil samples sourced from commercial markets in Mexico, the United States, Spain, and Australia. To validate their findings chemically, they used a comprehensive set of standards, including 37 fatty acid methyl esters (FAMEs), tocopherols (α, β, γ, δ), phytosterols, and derivatization reagents obtained from Sigma-Aldrich (USA) and Sinopharm (China).
Chemical profiling followed established techniques:
- FAMEs were analyzed using gas chromatography–mass spectrometry (GC-MS),
- Tocopherols and phytosterols were measured via high-performance liquid chromatography (HPLC), with slight protocol modifications,
- Adulteration markers such as isoflavones and oryzanol—characteristic of soybean and rapeseed oils—were identified using liquid chromatography–mass spectrometry (LC-MS) and reversed-phase HPLC (RP-HPLC).
Statistical modeling was conducted in MATLAB R2020a using OCPLS, with authenticity evaluated based on two key metrics: score distance (SD), which measures proximity to the class center in latent variable space, and ACR , which captures variability in the model’s projection. Samples with low ACR values were classified as authentic. Compositional differences were further analyzed using ANOVA with Tukey’s test (p < 0.05) in IBM SPSS v21.
By combining chemical fingerprinting with probabilistic modeling, the method offers a way to detect adulteration without the need to know in advance what the adulterant might be.
How It Works—and What It Found
The approach rests on a key principle: when models are built solely from authentic samples, their ACR sums tend to be higher. When adulterated samples are introduced, they distort the model’s boundaries, reducing ACR sums and making it harder to detect anomalies—unless you know what to look for.
To address this, the researchers ran the analysis in three main steps:
- Model generation: 20,000 OCC models were created using Monte Carlo sampling, each trained on a random subset (40–50 %) of the total data.
- Outlier detection: “Good” models—those with high ACR sums—were identified, and samples with an 80 % or greater probability of appearing in test sets across these models were flagged as potentially adulterated.
- Validation: Sampling ratios were adjusted to confirm the results, and flagged samples were subjected to chemical testing.
Out of the 40 samples, six (labeled 2, 3, 8, 23, 32, and 33) were identified as adulterated. Chemical analysis confirmed:
- Samples 2, 3, and 8 had elevated levels of α-linolenic acid, γ-tocopherol, and isoflavones—markers typically associated with soybean oil.
- Sample 23 showed increased erucic acid and brassicasterol, suggesting contamination with rapeseed oil.
- Samples 32 and 33 contained high levels of campesterol and oryzanols, indicating the presence of corn oil.
The strong agreement between the algorithmic predictions and chemical validation demonstrates the model’s ability to detect unknown adulterants—even in cases of low-level contamination—without relying on predefined training data.
Conclusion
This real-time OCC approach represents a practical, scalable solution for authenticating high-value edible oils. Unlike traditional models that rely on extensive training datasets or known adulterants, this method adapts to real-world variability and can be deployed even when the nature of adulteration is unknown.
By focusing on statistical behavior drawn exclusively from authentic samples, the approach simplifies implementation and improves inspection reliability—especially in market surveillance settings where time, resources, and prior data may be limited.
While this study focused on avocado oil, the framework could be extended to other food products vulnerable to economically motivated adulteration. Future research may explore integration with advanced spectral techniques, automated decision systems, or broader food matrices to further enhance fraud detection capabilities.
Journal Reference
Monti, M., Fiorentino, J., Dimitrios Miltiadis-Vrachnos, Bini, G., Tiziana Cotrufo, Sanchez de Groot, N., Alexandros Armaos, & Tartaglia, G. (2025). catGRANULE 2.0: accurate predictions of liquid-liquid phase separating proteins at single amino acid resolution. Genome Biology, 26(1). DOI: 10.1186/s13059-025-03497-7
https://genomebiology.biomedcentral.com/articles/10.1186/s13059-025-03497-7
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.