Geometric analysis and prediction systems for protein domain interacting interfaces

Abstract

One of the key mechanisms of biological functions is protein domain-domain interactions. Domain interactions and hence domain interfaces play a key role in the elucidation of cellular functions. Therefore, structure-based drug discovery will increasingly rely on a comprehensive understanding of domain interactions. For a comprehensive understanding of domain-domain interactions, large-scale geometric analyses on protein domain interfaces are crucial. Here, we propose three geometric analysis systems which address the problems of defining interface surfaces between protein domains, comparing and classifying interface structures, and predicting protein domain interfaces on newly discovered protein structures. To define interface surfaces accurately, we propose a watertight boundary detection algorithm that detects the watertight boundary on the interface surfaces between two domains defined by a Voronoi diagram. Using the proposed method, over 47,000 interface surfaces are extracted and stored to the InterPare database. Analysis of the interface surface dataset determines that the minimum interface area is 300 $Å^2$. In addition, we present an interface comparison method. Our comparison scheme using spin-image matching works well. And several 3D surface similarity measures are used for quantitative comparison of interface surfaces. By exhaustive N-by-N interface comparisons and hierarchical clustering analysis, we subclassed 10 sample SCOP families into 500 subfamilies. We found that in the subfamily not only interface structures but also interaction patterns are conserved over domains. We conclude by proposing a novel domain interface prediction method on newly discovered protein structures. Furthermore, prototype implementation using non-manifold spin-image matching predicts domain interfaces well using our unique interface prediction mechanism.