NEURAL-NETWORK APPLICATIONS IN THE SELECTIVE SEPARATION OF BIOLOGICALPRODUCTS

Neural networks have been applied to adsorptive separation, notably ch romatography, and aqueous two-phase separation. Phenomenological model s for them are either oversimplified or are too complex for easy desig n, scale-up and on-line implementation Adsorptive separations have bee n described by simple networks with topologies such as 3-2-3 and 4-4-1 . Aqueous two-phase separations are more complex. In a study of the re covery of selected proteins from a multi-component solution, a hierarc hical network with three subnetworks feeding two hidden layers of neur ons was needed. The network was flexible, easier to solve than some ph enomenological models, and could be integrated with an expert system f or on-line optimisation. The application of neural analysis to product recovery methods is an important component of bioprocess optimisation . The combination of neural networks with expert systems enables the d evelopment of integrated systems for optimal design and operation of f ermentation-cum-recovery plants.