Extraction of 2D-pattern signatures for protein folds

Protein contact map is a 2-dimensional representation of the protein tertiary structure. It is conjectured that proteins of the same SCOP structural fold give rise to `similar´ contact maps. The goal of this work is to investigate this hypothesis. We adopt data-driven approach by extracting conserved 2D-patterns within contact maps of proteins and validate the fact that many conserved patterns do correspond to structural elements within the corresponding protein tertiary structure, thus providing a potentially fast method for protein structure comparison. An efficient linear algorithm that extracts patterns from a contact map is implemented to create a 2D-pattern data base. Checking for redundancy is a time-consuming step and we apply an approximate 2D-pattern matching algorithm in order to obtain significantly distinct 2D-patterns. Using the framework of Class based association (CBA) rule mining, these patterns are then composed to form a `signature´ for a protein fold. Further, these 2D-features from contact maps are used successfully in addressing the challenging problem of protein fold classification.