Prediction of Secondary Structures of Proteins

Abstract

Artificial neural network techniques and Statistical techniques have been used for the prediction of secondary structures of proteins. Back-error propagation was used initially. However it has some inherent shortcomings in its implementation, one being a long convergence time for training and the other being the occurrence of local minima. To overcome the above mentioned drawbacks, a different algorithm called Fuzzy-Adaptive Resonance Theory Mapping (ARTMAP) was employed for our complex pattern mapping problem. Fuzzy ARTMAP is an incremental supervised learning algorithm which combines fuzzy logic and adaptive resonance theory neural network for the recognition of pattern categories.The Chou-Fasman(C-F) algorithm is a statistical procedure based on assigning conformation potentials to all the amino acid residues. The conformational potentials, one for each confirmation state, are obtained from statistical analysis of proteins of known secondary structure. A commercial package microgene, which implements the Garnier’s algorithm(statistical method) was also used in this research. Results of all the three methods are compared in this paper.The following conclusions have been drawn from this study: 1. The artificial neural networks can be employed to successfully predict the 3-dimensional structures of the unknown proteins and provide an alternative to the experimental techniques such as X-ray crystallography. 2. The neural network based on the supervised adaptive resonance theory provides a system which can make 3D structure predictions, comparable in accuracy to the back- propagation algorithm and in a much shorter time. 3. The statistical technique based on Chou-Fasman rules also gave good prediction results. But with the neural network technique the network can be improved by adding more proteins to it. Also, the lower percentage prediction in neural networks was observed in proteins having alpha-helixes which means results should improve after adding another 20–30 proteins having alpha-helixes in them.