Correlations between predicted and measured cell density (A, B) or antibody titer (C, D), in 96-well plates or bioreactor cultures, as indicated. Open circles correspond to calibration data points and full circles correspond to validation data points. The optimum number of latent variables was selected such that the root mean squared error observed in the validation data set was minimum. For cell density, the best model structures are composed by 6 or 7 latent variables (LVs), corresponding to average errors of 10% or 9%, in the microtiter plates or bioreactor cultures, respectively. Larger models of 9 LVs allowed the best description for antibody titer, with average errors below 7%. Contour plots show the relative fluorescence changes in the map during a culture, analyzed in (E) 96-well plates and (F) in bioreactors. Colors reflect the magnitude of the ratio between maps from the end of a CHO cell culture and the beginning of the same culture.