Evaluation of rapid evaporative ionization mass spectrometry (REIMS) for the prediction of slice shear force and quality grades in beef longissimus lumborum steaks.

Steak samples were collected from the longissimus lumborum muscles of beef carcasses (Canada AA, n = 1505; Canada AAA, n = 1363) over a 3-year period. Steaks were aged for 14 d, and tenderness was determined by slice shear force (SSF). Metabolomic profiling of beef samples was performed using rapid evaporative ionization mass spectrometry (REIMS) (N = 2853). Thirteen machine learning algorithms were used to build predictive models. Data were reduced using two separate approaches, one being feature selection (FS) and the second principal component analysis followed by FS (PCA-FS). No models could predict SSF tenderness category using FS and PCA-FS datasets with higher accuracy than the no information rate (NIR; 59.5 %, P ≥ 0.05). Population mean and standard deviation (SD) were calculated to generate 4 SD categories (±2) for further predictions. No model could predict SD category using the FS dataset (NIR = 55.1 %, P > 0.05). Top accuracies for PCA-FS were generated from the Treebag and Random Forest (RF) algorithms (82.8 % and 83.0 %, respectively; NIR = 55.0 %, P < 0.001). Top accuracies for FS were generated from SVM Radial and XGBoost to predict quality grade (84.6 % and 85.3 %, respectively NIR = 52.5 %, P < 0.001). Top accuracies for PCA-FS were generated from SVM Radial and RF (82.8 % and 84.2 %, respectively, P < 0.001). A stepwise regression model was built to evaluate relationships between SSF values and spectra generated from REIMS. Selected REIMS bins accounted for 7.2 % of the variation in predicted SSF values (R² = 0.072; P < 0.001). With more development, the RF algorithm could assist REIMS in rapid assessment of carcass quality.