Development of MLPs neural network and investigation of adaptive techniques in the network training for bioinformatics application

Abstract

In this paper, Multilayer Perceptrons (MLPs) neural network has been implemented in order to predict the protein secondary structure. The comparison of adaptive techniques in the network training also presented in this paper. The training of neural network is based on local and dynamic adaptive techniques and the training of each adaptive technique has been compared with respect to the convergence time. Besides, investigation was undertaken to verify the convergence time for these adaptive techniques. Based on the simulation results, RPROP is superior to the other adaptive techniques with respect to the convergence time in the protein secondary structure prediction while Delta Bar Delta rule seen to be perform the slowest training. Overall, Local adaptive techniques are perform faster than dynamic adaptive techniques in this case.