Protein Secondary Structure Prediction: A Feed-forward Neural Network approach

Protein secondary structure prediction is a very hot topic in bioinformatics. Predicting protein secondary structure means to find out the portions that contain Helix and Sheet in protein sequence. There are several methods for predicting protein secondary structure. The methods like Genetic Algorithm, Hidden Markov Model and different kinds of Neural Networks are there. Genetic Algorithm mostly deals with protein tertiary structure and sequence alignment, for Hidden Markov Model the accuracy is not good and Neural Network is the most successful for predicting protein secondary structure. So, we used the method named ?Feed Forward Neural Network? and implemented it with JOONE (Java Object Oriented Neural Engine) editor. At first we have classified the 20 protein according to their structure, size and hydrophobic manner. Then we have modeled a new architecture in feed forward network and used those classified proteins as input. Our achieved accuracy of helix prediction is 71% and sheet prediction is 65%. The result shows the improvement over previous works done in this regard. We hope that our work will be a future directive in this arena.