(571b) Optimization of Multi-Dimensional CHO Cell Culture Processes in Chemically-Defined Media Using a Nonlinear Experimental Design Approach

CHO cell culture in chemically-defined medium is generally a highly non-linear and multi-dimensional process with up to 100 categorical and numerical variables. Optimization of a chemically-defined medium generally applicable to multiple CHO cell lines can be complicated and time-consuming. A quality database representing the complex interactions among all the potential variables such as medium components (over 80) and process parameters (over 20) will be critical and can preferably be obtained using a highly efficient, nonlinear design of experiment (n-DOE) method. The n-DOE method can be applied to build such a representative and informative database to find optimal operating conditions for the higher desirability of single or multiple responses (e.g., volumetric productivity, titer, integrated viable cell density, viability, ammonium, lactate). The n-DOE method, based on radial basis function neural network and truncated genetic algorithm, can utilize information accumulated from any previous experimental runs and suggest new experimental conditions towards the local or global optimum of the complex CHO cell culture process. Specifically, deficient test technique was successfully applied to five cell lines for an exhaustive search of the essential/significant nutritional requirements of CHO cells from more than 80 potential medium components in a starting medium. To further dynamically eliminate insignificant variables possibly existing in this multi-dimensional problem, data mining techniques (e.g. decision tree analysis) have been applied to the informative database to identify critical variables and suggest optimal variable ranges for the next set of experimental design. It has been proven that the n-DOE method and data mining techniques are capable of achieving similar or even better optimum in less than half experimental runs that are necessary to traditional statistical design methods.