One of the most important methods to produce porcine interferon ? is
microbial fermentation. In the present study, recombinant Pichia pastoriswas
used. Broth?s antiviral activity is the key index of the expression level of
porcine interferon ?. Measurement of antiviral activity is a time-consuming
and difficult task, which makes the research and production work inconvenient
and uncertain. To solve this problem, multivariable regression and artificial
neural network were applied to predict the antiviral activity based on five
on-line variables (induction time, temperature, dissolve doxygen, O2uptake
rate and CO2 evolution rate) and two off-line variables (methanol consumption
rate and total protein concentration).Parameters of the multivariable
quadratic polynomial regression equation were estimate dusing least square
methods. Optimization of artificial neural network(ANN)was achieved by
back-propagation and genetic algorithm. Verified by test set, the
ANN optimized by genetic algorithm had the best predictive performance and
generalization. The sensitivity analysis showed that CO2evolution rate,
O2 uptake rate and methanol consumption rate were the most relevant factors
for model?s output, except for the antiviral activity?s own previous value.