A New Approach to Protein Structure Mining and Alignment

One of the largest areas of bioinformatic and data mining research has been in the protein domain. These efforts have included protein structure prediction, folding pathway prediction, sequence alignment, ab initio simulation, structure alignment, substructure detection and many others. Substructure detection is generally defined as the mining of a molecule’s 3D structure in order to find interesting/frequent domains. Sequence alignment involves determining the similarity of two (or more) protein molecules based on the how well their amino acid sequences “match.” There are potential pitfalls when trying solve both of these problems, however. In the case of substructure mining, focusing solely on structural information can lead to the discovery of biologically irrelevant substructures. With sequence alignment, the alignment results can vary greatly, depending on the substitution matrix used. In this paper we describe a method that combines the benefits of both substructure mining and sequence alignment in an attempt to determine the similarity between protein molecules. In the absence of biological information, our work will quickly and efficiently mine a protein molecule in order to determine frequent local structures. With the addition of biological sequence information, however, our algorithm provides a way to align proteins with similar local structures and sequence, yielding a global alignment between molecules. We present a novel structure mining/alignment algorithm as well as some additional work into a new clustering metric for amino acids based on several different physico-chemical properties. This metric is used with our alignment algorithm in order to provide a mechanism for globally aligning protein molecules.