Protein Super family Classification using Artificial Neural
Networks

Abstract

Classification, or supervised learning, is one of the major data mining processes. Pattern recognition involves assigning a label to a given input value. Protein classification is a problem of pattern recognition. The classification of protein sequences is an important tool in the annotation of structural and functional properties to newly discovered proteins. This protein super family classification is used in drug discovery, prediction of molecular functions and medical diagnosis. Many techniques can be implemented for classification tasks such as statistical techniques, decision trees, support vector machines and neural networks. In this work, feed forward neural networks approach is used. Neural networks have been chosen as technical tools for the protein sequence classification task because: The features that are extracted from protein sequences are distributed in a high dimensional space and they have got complex characteristics which make it difficult to satisfactorily model using some parameterized approaches; and the rules produced by decision tree techniques are complex and difficult to understand because the features are extracted from long character strings.

In this work, a comparative study of training feed forward neural network using the three algorithms – Back propagation Algorithm, Levenberg marquardt Algorithm and Back propagation Algorithm with genetic algorithm as optimiser is done. The efficiency of the three algorithms is measured in terms of convergence rate and performance accuracy.

Keywords: ANN (Artificial neural network), Back propagation algorithm, Levenberg marquardt algorithm, Genetic algorithm.