The selection of medium feed rates for perfusion bioreactors represents a c
hallenge for process optimization, particularly in bioreactors that are sam
pled infrequently. When the present and immediate future of a bioprocess ca
n be adequately described, predictive control can minimize deviations from
set points in a manner that can maximize process consistency. Predictive co
ntrol of perfusion hollow-fiber bioreactors was investigated in a series of
hybridoma cell cultures that compared operator control to computer estimat
ion of feed rates. Adaptive software routines were developed to estimate th
e current and predict the future glucose uptake and lactate production of t
he bioprocess at each sampling interval. The current and future glucose upt
ake rates were used to select the perfusion feed rate in a designed respons
e to deviations from the set point values. The routines presented a graphic
al user interface through which the operator was able to view the up-to-dat
e culture performance and assess the model description of the immediate fut
ure culture performance. In addition, fewer samples were taken in the compu
ter-estimated cultures, reducing labor and analytical expense. The use of t
hese predictive controller routines and the graphical user interface decrea
sed the glucose and lactate concentration variances up to sevenfold, and an
tibody yields increased by 10% to 43%. (C) 1999 John Wiley & Sons, Inc.