关键词: 近红外建模, 偏最小二乘法, 聚苹果酸, 单因素培养基优化, 诱变菌种筛选, 配对t-检验

Abstract: High performance liquid chromatography (HPLC) was used to measure the polymalic acid concentration in the fermentation broth of Aureobasidium pullulans. Interval partial least square regression (iPLSR) combined with moving window PLSR (MWPLSR) was used to confirm 5 638–6 024 cm-1 as the waveband for modeling. Multiplicative scatter correlation (MSC), standard normal variate (SNV), Savitzky-Golay 55 points smoothing and 1st derivative spectrum were successively operated as spectral pre-processing methods, and then PLSR with the first 5 factors was used to develop a predictive model with the highest accuracy. The root mean square of prediction (RMSEP) and correlation coefficient of prediction (Rp) of the model were 1.553 g/L and 0.970 0 for the internal test set, and 1.378 g/L and 0.992 4 for the external test set, respectively, and a paired t-test at 95% confidence level demonstrated that the maximum deviations between the HPLC values and the model predicted values were calculated as 1.48 and 0.83 g/L, respectively. There was no significant difference among three parallel predictions for one sample. When the model was applied to medium composition optimization and mutant screening, large prediction errors were found with paired t-test. However, the model could reliably predict polymalic acid concentration under the premise that the model calculated values were larger than 3.19 and 1.44 g/L, respectively, indicating the potential application of the near-infrared model in medium composition optimization and strain screening.

Key words: near-infrared modeling, partial least square, polymalic acid, medium optimization, strain screening, paired t-test