MULTIGRADIENT METHOD FOR OPTIMIZATION OF SLOW BIOTECHNOLOGICAL PROCESSES

A new method (named a ''jumping spider'') is introduced for the optimi zation of slow biotechnological processes. The more traditional sequen tial experimentation (i.e., gradient search, simplex, etc.) is not wel l suited for slow dynamic processes, e.g., plant cell culture and diff erentiation. Therefore, a more simultaneous approach is proposed. A la rge number of initial experiments are performed, on the basis of which several of the initial experiments are selected as starting points. A search is then performed simultaneously from several gradient directi ons and the optimum is estimated by a quadratic approximation. In simu lations, the spider generally climbs up the slopes quickly and the fin al estimator yields good maximum point estimates even on a complex top ography. The spider may even approach more than one local maximum poin t simultaneously. As a model application, the average xylitol conversi on rate,of Candida guilliermondii was optimized in relation to cultiva tion volume (oxygen availability) and the concentration of nitrogen an d phosphorus in the medium. A threefold increase in xylitol production was obtained with three experimental steps. (C) 1993 John Wiley & Son s, Inc.