The Research On The Application Of Artificial Neural Network In The Predication Of The Secondary Structure Of Protein

With the rapid increase of newly discovered primary structure of protein, secondary structureprediction by automated methods is urgently needed. Significant progress has been achieved in thisaspect such as the development of Artificial Neural Networks (ANNs). Besides its high workingefficiency, ANNs still suffers from an inadequacy of the accuracy in prediction. In order to solve thisproblem, an ideal predicting engine, as well as an efficient encoding algorithm is required. In this case,more useful information from the primary structure will be extracted to generate an all-around modelfor the secondary structure.This paper describes the current progress and significance of the researches on protein secondarystructure prediction, accompanied by emphasis on the problems of the existing encoding algorithms.Improvement on Profile encoding algorithm has been made to maximize the efficiency of extractingevolutionary information from existing amino acid sequence. More importantly, a new encodingalgorithm has been designed based on the analysis of the Profile version to achieve a higher accuracy,which has been tested for implementation.We first employ BP Neural Networks which can slove the non-linear mapping function as thepredicting engine and used two common encoding algorithms (Orthogonal encoding algorithm andProfile encoding algorithm) respectively to predict the secondary structure. Then we improved Profileencoding algorithm by considering the long-range interaction information between amid acids. Wetest the improved encoding algorithm using the BP Neural Network as the predicting engine, whichshows an increase of3%in accuracy compared to common encoding algorithm.