Performance evaluation of protein structural class prediction using artificial neural networks

Prediction of protein structural class has been a new area of research in the scientific community in the last decade. Various approaches has been adopted and analysed. However representing the raw amino acid sequence to preserve the property of proteins has posed a great challenge. Chou´s pseudo amino acid composition feature representation method has fetched wide attention in this regard. In Chou´s representation each protein molecule is represented as the combination of amino acid composition information, the amphiphillic correlation factors and the spectral characteristics of the protein. This method preserves both the sequence order and length information of the raw amino acid sequence and this plays a significant role in prediction. A set of exhaustive simulation studies with functional link artificial network(FLANN) demonstrates high success rate of classification. The self-consistency and jackknife test on the benchmark datasets has been performed and a comparison has been done with the results of radial basis function (RBF) neural network. It indicates that the FLANN model´s accuracy is little less than RBF, but its complexity is very less whereas the accuracy of RBF is little higher, but it´s complexity is high in comparison to FLANN.