BACKGROUNDIn mammalian cell culture processes, viable cell density (VCD) is one of the most crucial parameters for monitoring of the current status of cell growth and protein production. Among the various real-time monitoring tools available for VCD in Chinese hamster ovary (CHO) cell culture, dielectric spectroscopy is one of the more reliable due to its simplicity, robustness and accuracy. These attributes enable efficient monitoring and control of the corresponding processes. However, one of the major drawbacks of dielectric spectroscopy is frequent deterioration of accuracy during the late-growth phase of CHO cell culture. This can be caused by abrupt changes in the cell physiology, as well as environmental conditions. RESULTSIn this study, various kinds of chemometric tools were employed to understand the spectral dynamics of dielectric spectroscopy during extended CHO cell culture, including different cell growth phases. Then, the performance of different multivariate regression models were compared with conventional Cole-Cole equation-based approaches. The models were compared in their ability to estimate VCD in CHO cell culture from dielectric spectra from several batches of bioreactor operations. Multivariate analysis of dielectric spectral datasets clearly revealed the time-varying evolution of spectral dynamics during the transition between cell growth phases. The best prediction performance from the dielectric spectroscopy was realized using a regression model based on locally-weighted partial least squares (LWPLS). This estimation model best compensated for spectral dynamics observed during the different cell growth phases. CONCLUSIONDielectric spectroscopy combined with LWPLS allows achievement of optimal and stable estimation of VCD during CHO cell culture, regardless of inherent differences in cell growth phases. (c) 2014 Society of Chemical Industry