Recursive Neural Networks Based Recognition of Protein Folds on Account of Residue Correlation within Protein Sequences

Abstract

“Protein sequences can be regarded as slightly edited random strings”1. We applied the methods of estimating residue correlation within the protein sequences. First we use the mutual information (MI) of adjacent residues, and improve the long range correlation between nonadjacent residues by defining the mutual information vector (MIV) of each protein sequence. The correlation is based on residue hydropathy rather than protein-specific interaction. Like this we calculated the MIV of each protein sequence, and these MIV are further give to recursive neural network to obtain the classification of protein sequence. The modeling power of MIV was shown to be significantly better, reaching the level where proteins can be classified without alignment information.