An Approach for Protein Secondary Structure Prediction Using Neural Network

Prediction of tertiary structure of protein and the function of a given protein as efficiently, it is necessary to know the secondary structure of protein, however it is a critical need in biological science. The final destination of this research is to measure the performance of predicting secondary structure of protein by using Neural Network (NN) that inputs are primary sequences of protein. There are two phase in Neural Network, one is training phase that are used to learn the network, to recognize the relation between primary structure and their corresponding secondary structure on a sample set of 75 proteins (16671 residues) of that are known as secondary structure. In testing phase where used on 5 proteins (381 residues) primary sequences that try to predict corresponding secondary structure. In our approach using Neural Network approach, we use one hidden layer with no hidden unit or 25,50,75,100,125 hidden units and we also use different window size (1,2,3,5,7,9…21) to find maximum output. Maximum predicative accuracy of Neural Network is 71.89% at window size 17 and 75 hidden unit of hidden layer for three states helix (H), strand (E) and coil(C). We say that presented approach in this paper is simple with better time complexity in comparison to Jaewon’s work[4].