Optimization of phenyllactic acid biosynthesis and separation by machine learning with neural network and overlay sampling uniform design

Abstract

Machine learning methodology with neural network models was developed using the datasets based on the overlay sampling uniform design (OSUD) for optimization of phenyllactic acid biosynthesis and separation processes by cryogels. Compared with the multiple regression, the machine learning models exhibited a significant improvement of predictive accuracy of phenyllactic acid biosynthesis, in which the radial basis function neural network (RBFNN) model had the best predictive performance with the accuracy increased by 65.2 %. The combination of RBFNN and OSUD was further employed to optimize the chromatographic separation of phenyllactic acid from crude fermentation broth using two poly(hydroxyethyl methacrylate) based anion-exchange cryogel packed-beds (grafted with (vinylbenzyl)trimethylammonium chloride and N,N-dimethylaminoethyl methacrylate). After optimizing the three critical separation parameters: sample volume (5.3–31.8 mL), flow velocity (1.0–6.0 cm/min), and elution salt concentration (0.05–0.3 mol/L), it was found that the models provided excellent predictions. The optimized recovery rates for the two packed-beds were determined to be 76.5 % and 83.0 %, and the optimal adsorption capacities were 0.26 mg/mL and 0.39 mg/mL from the fermentation broth, respectively. This study provides a reliable integrated approach for optimizing the synthesis and separation processes of high-value bioproducts like phenyllactic acid from crude feedstocks.