Proteomics coupled machine learning-innovative approach in geographical origin authentication of green Coffea arabica

The geographical authentication of green specialty coffee is an economically sensitive analytical task that is not yet fully resolved. We used an innovative combination of proteomic profiling with linear discriminant analysis for the authentication of the geographical origin of green specialty coffee beans from well-known harvesting regions in Central America, South America, Africa, and Asia. Out of 1596 identified proteins, we selected the top 30 target markers ranked by ANOVA. The model's prediction performance using leave-one-out cross-validation reached 85.3 %, with the lowest accuracy in the prediction rate for Asian samples. Besides, model performance and prediction sensitivity to random states were tested using 5-fold cross-validation. After 20 iterations, the model performance slightly decreased to 84.0 %. This research contributes to advancing traceability tools in the coffee industry, ensuring product authenticity, and promoting fair trade practices. Specificity and sensitivity confirmed that the model appears to be reliable at distinguishing Asian and African samples.