ENHANCED SUPERVISION OF RECOMBINANT ESCHERICHIA-COLI FERMENTATIONS VIA ARTIFICIAL NEURAL NETWORKS

This paper describes how artificial neural networks may be used to aid the supervision of recombinant E. coli fermentations. Two particular supervision aspects are considered. First, on-line estimators for biomass and recombinant protein concentration were constructed using information available on-line. Feedforward artificial neural networks, modified to include dynamic characteristics, form the basis of the estimator models. The estimators were developed for one particular fermentation system and the 'transferability' aspects of the procedures assessed by considering application to two different systems. Secondly, auto-associative neural networks were investigated for data feature extraction. The objective, to obtain an on-line indicator of plasmid structural instability, was achieved. Results are presented from industrial fermentation laboratory studies.