(A) study on real-time adaptive optimization of fed-batch culture of recombinant yeast by using artificial intelligence

Abstract

An on-line optimization algorithm for fed-batch cultures of recombinant yeasts to determine the optimal substrate feed rate profile has been developed. Its development involved four key steps: (1) development of reliable process model, (2) development of optimization algorithm, (3) design of on-line model update algorithm to be incorporated into the optimization algorithm and (4) ex-perimental validation. Artificial intelligence technique was used in each step to effectively handle the complex and nonlinear systems with a high computa-tional efficiency which was essential for on-line purposes. A recombinant Sac-charomyces cerevisiae producing human parathyroid hormone (hPTH) was chosen as the model strain. Three optimization strategies (GA I, GA II, and GA III) based on genetic algorithms were developed in the simulation study. The performances of those three algorithms were compared with one another and with that of a variational calculus approach. GA III, in which the length of feeding intervals and the feed rate in each interval were simultaneously optimized showed the best perform-ance and was selected to be used in the subsequent experimental study. To confer an adaptability to GA III, an on-line model update algorithm $(GA_MU)$ was developed and incorporated into GA III. The resulting on-line op-timization algorithm was experimentally applied to fed-batch cultures of re-combinant yeast with the same promoter and a similar plasmid size to the model strain, to maximize the cell mass concentration. It followed the actual process quite well and gave a 2.4-fold higher cell mass concentration than the open-loop control strategy, GA III with no model update by $GA_MU$. A linguistic process model based on self-organizing genetic-fuzzy system was developed in order to overcome the major weakness, a structural inaccu-racy in using a deterministic mathematical model. Attractive features of the model include that its structure is generated by itself and that the s...