Prediction of domain boundaries in protein sequences using predicted secondary structure and physicochemical properties of amino acids

The domain boundary prediction is an important task for functional classification of proteins, a homology based protein structure prediction and the high throughput structural genomics. A technique is proposed to predict protein domain boundaries using predicted secondary structure solvent accessibility and physicochemical properties of Amino Acids along protein chain. Each amino acid is represented by a set of physicochemical properties from AAIndex database. Support Vector Machine is used as a two class classifier where maximum prediction accuracy is observed by varying kernel functions and appropriate parameters. To improve the accuracy of prediction of multiple domains in protein chains, a Support Vector Machine is used as the classifier and physicochemical properties of proteins are used in this paper. It is tested on target proteins and achieves significant precision and recall.