Neural Network: A Machine Learning Technique for Tertiary Structure Prediction of Proteins from Peptide Sequences

The current work has deduced the novel method for tertiary structure prediction of various important unpredicted proteins through machine learning technique neural network. Multi-layer perceptron architecture has been developed to predict the tertiary structure (Phi/Psi) of proteins. A novel binary codification system has been devised for input and output processing. Twenty physiochemical properties representation scheme for each amino acid and binary output discretization of real valued torsion angle for each angle of residues has been adopted. The proposed system has been tested with different number of neural networks, training set sizes and training epochs. The overall successful prediction of residues for tertiary structure prediction (Phi/Psi) of protein has been reported according to window size as 9(52.4% / 56.2%), 13(56.5% / 61.3%), 17(53.7% / 57.2%), 21(53.2% / 57.4%). This study demonstrated the prospect of implementing fast and efficient structure prediction of peptide sequences using neural network.