Abstract
The classical approach to optimize biotechnological processes is timeconsuming and allows only a single-dimensional search therefore statistical-mathematical methods represent a great improvement in experimental design. Among these methods genetic algorithms (GA), which mimic the natural evolution, seem to be promising. In our laboratory the method of GA coupled with computer learning was used for medium optimization during the study of a fungal secondary metabolite production. At the beginning 11 medium components and their minimal and maximum concentrations were selected according to literature. In the first generation of the GA 12 different fermentation experiments were done, each of them representing one "chromosome". "Genes" of the "chromosome" coded the concentrations of a particular component. Inductive learning was used to obtain operational knowledge base from the set of past experiments. This base was used as a background knowledge for GA. Four generations of fermentation experiments were performed. In the first two generations a broad parameter space was explored while in the following generations in the most promising narrow area was examined. According to the strategy used, in the first two generations only a limited number of fermentation experiments gave better results that the initial control experiment while in the two last generations nearly all of the results were improved. During such optimization the yield of the desired secondary metabolite was nearly 5 times increased what could not be reached using classical optimization methods.