Development of Soft Sensors for Online Biomass Prediction in Production of Hepatitis B Vaccine

Abstract

Neural networks can provide highly accurate and robust solutions for complex non-linear kinetic like bioprocess reactions. It is applied as a soft sensor in the biotechnology industry for measuring and analyzing parameters that can hardly be reached online, as well as the reproducibility of the product without significant deviation. This study attempted to obtain the best neural network structure for online estimation of P. pastoris yeast biomass, which is used to express the hepatitis B surface antigen (HBsAg). During the fed-batch fermentation process, the CO2 evolution rate (CEO), ammonia consumption rate (ACR), and methanol consumption rate (MCR) were considered as inputs and biomass (WCW) as ANN output parameter. The results showed that after training the neural network structure using 4 fed-batch fermentation batches performed in the laboratory, biomass estimation (WCW) was obtained with high accuracy at specific times of the fed-batch fermentation process. R-Squared and RSME between actual and ANN estimations were 0.999 and 9.57, respectively.