Analyzing the Impact of Protein Representation on Mining Structural Patterns from Protein Data

Abstract

Proteins have essential roles in the biological processes of living organisms by contributing to maintain cellular environments. Understanding the conformational transitions of proteins may help identifying situations when incorrect folding or mutations can occur and thus, it may contribute to inhibit possible uncontrolled behaviour. In this paper we are performing a study on how different protein representations impact the process of mining relevant patterns from protein related data. Two representations are used for the proteins, one using the structural alphabet and the second using the relative solvent accessibility values of the amino acids from the proteins' primary structure. Using these representations, two case studies are performed to emphasize the effectiveness of using the proposed protein representations to unsupervisedly learn structural patterns from on a protein data set.