Optimizing the Generalization Ability of Artificial Neural Networks in ELISA Protocols by Employing Different Topologies and GENETIC Operators

The aim of the present work was to present the ability of Artificial Neural Networks (ANN) in successfully predicting the response of an Enzyme-linked Immunosorbent assay (ELISA) from the relative input parameters and to further enhance its performance by use of different network architectures, learning algorithms and genetic operators. Representatives of three major categories of ANN topologies were investigated, namely Multilayer Feed-Forward (MLF), Generalized Feed-Forward (GenFF) and Radial Basis Function (RBF) which were trained with back-propagation with momentum and a scaled conjugate gradient learning algorithm. Tuning of the input and hidden layer size was performed by use of a genetic algorithm, while different combinations of genetic operators were used for the optimal GenFF network in order to increase its predictive ability. The major advantage of this approach was the simultaneous data-driven, modeling and optimization process which demands no a priori knowledge of variable correlations and can be employed in setting up new assays.