BACKGROUNDIn recent decades, artificial neural network (ANN) has been shown to be a robust and promising tool in monitoring and controlling bioprocess systems. In a previous study, the authors designed a highly accurate and precise recurrent neural network (RNN) for predicting the biomass amount of recombinant Pichia pastoris Mut(+) producing intracellular hepatitis B surface antigen (HBsAg) during fed-batch methanol fermentation. In the current work, the aim was to compare the production efficiency of HBsAg between conventional predefined mu -stat methanol feeding control (open-loop control) - already established in large-scale production - and methanol feeding based on a mu -stat feedback control system using RNN. For this purpose, for each methanol feeding strategy, bench-scale, fed-batch fermentation processes were carried out twice. RESULTSAccording to the results, in contrast to the established mu -stat predefined feeding strategy (open-loop), the deviation of specific growth rate and biomass in mu -stat feedback control based on RNN was negligible. Also, in the suggested methanol feeding control strategy, the HBsAg titer, specific productivity and yield between the performed fed-batch fermentations - unlike the conventional method - were approximately identical, with average values of 110.8 mu gmL(-1), 1.52 mu gg(-1) dry biomass and 0.34 mu gg(-1) MeOH respectively. CONCLUSIONComparing the biomass growth pattern and HBsAg production efficiency with the conventional mu -stat predefined feeding in open-loop control, the new proposed feeding control system illustrated significantly high process efficiency. This reliable control system based on ANN can have many applications in the biopharmaceutical industry for the control of process key parameters as well as for enhancing process efficiency. (c) 2019 Society of Chemical Industry