Geographical origin authentication of lamb: Advancing label control through 4D-DIA quantitative proteomics and interpretable machine learning methods

The authenticity of food origin labelling is a critical concern for consumers. This study analysed the protein profiles of Tan lambs from three different regions using 4D-DIA quantitative proteomics. Using least absolute shrinkage and selection operator (LASSO), 16 potential protein markers were identified. Six machine learning algorithms were employed to train predictive models, with leave-one-out cross-validation. Significant regional differences in protein composition and abundance were revealed by the results. Fourteen protein markers selected using λ min demonstrated superior predictive accuracy compared to the λ 1se group, likely due to its reduced information loss. Logistic regression was identified as the optimal model based on the receiver operating characteristic curve performance. Further analysis using the SHapley Additive Explanations method highlighted W5PF65, W5PQE5 and W5Q501 as key markers for origin identification. This study provides an in-depth analysis of the protein composition of Tan lamb across different regions. It also presents a method for verifying the origin of lamb products and enhancing food traceability.