Optimization of Lipase-Catalyzed Synthesis of Vitamin A Ester (Retinyl Laurate) by Response Surface Methodology and Artificial Neural Network

• Main Authors: Hsin-Ju Li, 李心汝
• Other Authors: Yung-Chuan Liu
• Format: Others
• Language: zh-TW
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/55670010306261613088

碩士 === 國立中興大學 === 化學工程學系所 === 104 === In this study, immobilized lipase from Candida antarctica (Novozym® 435) was used in enzymatic catalysis to generate retinyl laurate by esterification of retinyl acetate with lauric acid in n-hexane. This reaction was model and optimized by response surface methodology (RSM) and artificial neural network (ANN). A 3-level-4-factor central composite design (CCD) was employed in this study to design 27 experiments. The experimental variables included enzyme amounts (10~50 mg), temperature (40~60℃), time (2~6 hours) and substrate molar ratio (acid/alcohol=1~5). Within the 27 experiments, the highest and lowest relative conversion are 82.64 ± 1.94 and 14.52 ± 1.85%, respectively. The coefficient of determination (R2) calculated from the design data of RSM is 0.9887, and analysis of variance (ANOVA) showed that enzyme amounts and substrate molar ratio exhibited the stronger effect on the molar conversion than other variables. As ANN was employed to model the biosynthesis, the optimal architecture of ANN was founded as follows: learning algorithm (LM), transfer function (Tanh), iterations (104) and the nodes of hidden layer (6). The root mean squared error (RMSE) and R2 were 0.22347 and 0.99994, respectively, suggesting that ANN was better than RSM in data fitting. The relative conversion of RSM and ANN optimal verification test were 84.7 ± 1.16% and 88.31 ± 0.3%, and the relative conversion decreased obviously when reusing enzymes by the fifth times. The results of verification test suggested that the RMSE of RSM and ANN were 0.04 and 0.12, respectively. This result shows ANN is more appropriate as the system of data prediction in this study.