Optimization of artificial neural network by genetic algorithm for describing viral production from uniform design data

• We report an artificial neural network (ANN) for modeling rabies virus production.
• A genetic algorithm (GA) approach was used to optimize ANN topology.
• A general model based on ANN was defined as a correlation technique for obtaining uniform experimental design (UD) data.
• Results derived from this work can be useful for developing viral vaccines and process staff in pharmaceutical organizations.

This work objective was to define a modeling approach based on genetic algorithm (GA) for optimizing parameters of an artificial neural network (ANN); the latter describes rabies virus production in BHK-21 cells based on empirical data derived from uniform designs (UDs) with different numbers of experimental runs. The parameters considered for viral infection were temperature (34 and 37 °C), multiplicity of infection (0.04, 0.07, and 0.1), infection, and harvest times (24, 48, and 72 h), with virus production as the monitored output variable. A multilevel factorial experimental design was performed and used to train, validate, and test the ANN. Its experimental fractions (18, 24, 30, 36, and 42 runs) defined by UDs were used to simulate the neural architectures. In GA, the neural computing parameters constituted the population individuals, and the steps involved were population creation, selection, and replacement by crossover and mutation. The ANN optimized by the combined algorithm showed a good calibration for all UDs under consideration, thus demonstrating to be suitable (R > 0.85) as a correlation method in UDs independent of the experimental runs developed. Therefore, this work could guide researchers in the efficient use of UDs in the simulation and optimization of virus production processes.

Figure optionsDownload as PowerPoint slide