Feature Extraction of Protein Contact Maps from Protein 3D-Coordinates

This work mainly proposes an alternate way of solving challenging problems of computational biology like protein secondary structure assignment, protein fold identification/recognition, protein fold signatures, and contact map overlap problem by exploiting the idea that proteins belonging to the same protein fold have similar contact maps. Pattern mining of contact maps is conducted to extract features that pertain to fold information. Using the work in the literature that predicts contact maps from the primary amino acid sequence, we propose that using pattern features from predicted contact maps would lead to an Ab-Initio method. Hence, instead of extracting features from the primary amino acid sequence, we propose to extract pattern features from the protein contact maps. Protein secondary structure assignment is achieved with an accuracy of 76% on RS126 data set, on par with best of algorithms up to 10% of noise, and then the performance falls to 66% by 15% noise.