Protein Secondary Structure Graphs as Predictors for Protein Function

Abstract

Predicting the functions of the proteins from their structure is an active area of interest. The current trends of the secondary structure representation use direct letter representation of the specific secondary structure element of every amino acid in the linear sequence. Using graph representation to represent the protein sequence provides additional information about the structural relationships within the amino acid sequence. This study outlines the protein secondary structure with a novel approach of representing the proteins using protein secondary structure graph where nodes are amino acids from the protein sequence, and the edges denote the peptide and hydrogen bonds that construct the secondary structure. The developed model for protein function prediction Structure2Function operates on these graphs with a defined variant of the present idea from deep learning on non-Euclidian graph-structure data, the Graph Convolutional Networks (GCNs).