Protein Structure Similarity Measurement by Language Modeling Techniques

In the era of structural biology, it is necessary to apply efficient and effective tools in order to measure structural similarity between proteins. Although a great number of structural comparison methods have been developed, none of them gives an exact solution to the problem. In this paper, we introduce a novel method for structural similarity measurement of proteins based on language modeling techniques. The roots of the method are inspired from computational linguistics and the related techniques for quantifying and comparing strings of characters. In this way, the protein structure is represented in sequences of characters and then n-gram based modeling techniques are applied to capture the content regularities. In the sequel, these regularities are contrasted by cross-entropy concept and the similarity between two protein structures is measured. To find an overlap between two protein structures in 3D-space, a superposition task is also applied. In this very first attempt, the experimental results represent the usefulness of the new approach and motivate further studies on development of tools based on computational linguistics methods.