Monitoring Tremella fuciformis submerged fermentation using ATR-MIR combined with chemometrics

Abstract

This study employed mid-infrared attenuated total reflection (ATR-MIR) spectroscopy and chemometrics to monitor the submerged fermentation process of Tremella fuciformis (T. fuciformis). It investigated the effects of four different preprocessing methods on the performance of both the qualitative identification and the quantitative prediction models. The qualitative model, which employed unsupervised learning via Principal Component Analysis (PCA), analyzed ATR-MIR data, physicochemical parameters, and rheological parameters, clearly delineating distinct fermentation stages. The supervised Random Forest (RF) model optimized input variables through feature importance selection and PCA, achieving a classification accuracy of 97.5%. The quantitative model, Partial Least Squares Regression (PLSR), demonstrated strong predictive performance for reducing sugar, total sugar, tremella polysaccharide, and dry cell weight, with low root mean square error and high R² values. This ATR-MIR spectroscopy-based chemometrics model offers valuable insights for food science and holds the potential for optimizing tremella polysaccharide production through precise fermentation control.