Quantitative prediction of freshness of fish muscle tissue by machine learning coupled with MALDI-TOF MS

This study developed a prediction model for freshness analysis of fish muscle tissue based on MALDI-TOF mass spectrometry (MS) combined with machine learning (ML) algorithms. Dorsal muscle samples of large yellow croaker (Larimichthys crocea) stored at 4 °C for different periods were analyzed as models. The sensory quality of the samples was evaluated at 0, 1, 3, 5, and 7 days of storage using a 10-point scale. Sensory evaluation results show that the overall acceptability scores decreased from 9.8 at day 0 to 3.3 at day 7 during refrigerated storage, falling below the acceptance threshold (5.0) between day 3 and 5. The mass spectral data of protein extracts of dorsal fish muscle samples were obtained by MALDI-TOF MS. Six ML models were constructed based on 315 mass spectra, and compared. The XGBoost model shows the superior performance for predicting the freshness of large yellow croaker fish dorsal muscle tissue, with a coefficient of determination (R²) of 0.978, and residual predictive deviation (RPD) of 6.742. An 8:2 train/test split of mass spectral data is appropriate for constructing the models. The impact of data partitioning on the models' performance was evaluated using a leave-one-fish-out cross-validation (LOFO). XGBoost remains the optimal performing model (R² = 0.905 ± 0.074) under LOFO. A model performance ranking keeps consistent with that under the 8:2 split. SHAP analysis, combined with tryptic digestion and peptide mass fingerprinting searching, identified four proteins that can serve as potential biomarkers for monitoring freshness change during storage of the analyzed fish muscle tissue. During the single-blind validation using commercial samples, 29 predictions among the 30 analyzed samples remain within the model's error tolerance of 0.5 d, with an overall prediction accuracy of 96.67% (n = 30). It provides a valuable methodological reference for the laboratory-based quality assessment of aquatic products.