Characterization, classification and alignment of protein-protein interfaces

Abstract

Protein structural models provide essential information for the research on protein-protein interactions. In this dissertation, we describe two projects on the analysis of protein interactions using structural information. The focus of the first is to characterize and classify different types of interactions. We discriminate between biological obligate and biological non-obligate interactions, and crystal packing contacts. To this end, we defined six interface properties and used them to compare the three types of interactions in a hand-curated dataset. Based on the analysis, a classifier, named NOXclass, was constructed using a support vector machine algorithm in order to generate predictions of interaction types. NOXclass was tested on a non-redundant dataset of 243 protein-protein interactions and reaches an accuracy of 91.8%. The program is benecial for structural biologists for the interpretation of protein quaternary structures and to form hypotheses about the nature of proteinprotein interactions when experimental data are yet unavailable. In the second part of the dissertation, we present Galinter, a novel program for the geometrical comparison of protein-protein interfaces. The Galinter program aims at identifying similar patterns of different non-covalent interactions at interfaces. It is a graph-based approach optimized for aligning non-covalent interactions. A scoring scheme was developed for estimating the statistical signicance of the alignments. We tested the Galinter method on a published dataset of interfaces. Galinter alignments agree with those delivered by methods based on interface residue comparison and backbone structure comparison. In addition, we applied Galinter on four medically relevant examples of protein mimicry. Our results are consistent with previous human-curated analysis. The Galinter program provides an intuitive method of comparative analysis and visualization of binding modes and may assist in the prediction of interaction partners, and the design and engineering of protein interactions and interaction inhibitors.