ABSTRACT
Currently, the prediction of a three-dimensional protein structure from a protein sequence poses insurmountable difficulties. As an intermediate step, a much simpler task has been pursued extensively: predicting one-dimensional strings of secondary structure. Here, a composite neural network is described which predicts three secondary-structure states (helix, strand, loop). The network system comprises two levels of feedforward networks (one hidden layer each) and a final jury decision over differently trained networks. Training is done by an adaptive-like backpropagation. An important key feature of the system is that the input is not only the sequence of one protein but the profile of a set of sequences from proteins which have the same three-dimensional structure. The combination of the problem-specific topology and the preprocessing of the input improve prediction accuracy from 62% to 72%. Furthermore, the specific topology and training procedure successfully correct for shortcomings of both simpler neural network and classical methods. Over the last few years, the network system has been the best automatic predictor in a very competitive area of research.
