Modeling bioprocess scale-up utilizing regularized linear and logistic regression

Bioprocess scale-up from optimized flask cultivations to large industrial fermentations carries technical challenges and economical risks. Essentially, the prediction of optimal process conditions in large fermentations based on small scale experiments is non-trivial. For example, common statistical methods encounter problems with the high-dimensional, small sample size and, on the other hand, the use of various scale-up criteria requires a priori knowledge that may be difficult to obtain. We propose a novel computational scale-up approach applicable to various bioprocesses. The method bases on regularized linear and logistic regression. With embedded feature selection, it automatically identifies the most influential parameters and predicts their values in large scale. In addition, the method predicts the resulting large scale yield. As a case study, we examined the production of a cytotoxic compound. We predicted scale-up from flask and 2L to 30L fermentations and found that, in both cases, the product yield predictions are close to experimentally observed yields.