Addition of Contact Number Information Can Improve Protein Secondary Structure Prediction by Neural Networks

Prediction of protein secondary structures is one of the oldest problems in Bioinformatics. Although several different methods have been proposed to tackle this problem, none of these methods are perfect. Recently, it is proposed that addition of other structural information like accessible surface area of residues or prior information about protein structural class can significantly improve the prediction of secondary structures. In this work, we propose that contact number information can be considered as another useful source of information for improvement of secondary structure prediction. Since contact number, i. e. the number of other amino acid residues in the structural neighbourhood of a certain residue, depends on the secondary structure of the residue, we conjectured that contact number data can improve secondary structure prediction. We used two closely related neural networks to predict secondary structures. The only difference in the neural networks was that one of them was also provided with residue contact numbers as an additional input. Results suggested that addition of contact number information can result in a small, but significant improvement in prediction of secondary structures in proteins. Our results suggest that residue contact numbers can be used as a rich source of information for improvement of protein secondary structure prediction.