BACKGROUNDA goal in the production of biopharmaceuticals is to replace the cost-intensive, empirical quality by testing' approach with rational, knowledge-based quality by design' concepts. The major challenges in this context are the complexity of bioprocesses and the limited online access to process variables related to product quality. The implementation of advanced monitoring strategies combined with chemometric-based approaches represents a strategy to overcome this bottleneck. RESULTSA series of recombinant E. coli fed-batch production processes was conducted to provide an accurate data set for development of predictive statistical models. The applicability of partial least squares regression and radial basis function artificial neural network-based models to predict gene dosage, product titer, and solubility of the target protein was evaluated. In addition to signals from standard online measurements, multi-wavelength fluorescence signals turned out to provide essential information for prediction of product titer and quality. CONCLUSIONSThe application of advanced process monitoring concepts comprising recently established analytical targets and statistical modeling allows for real-time prediction of product related not directly accessible process variables. Availability of such process information facilitates design of advanced process-control strategies and will strongly support the implementation of quality by design in biopharmaceutical production. (c) 2014 Society of Chemical Industry