Docking Benchmark Set of Protein Models

Abstract

In structure-based genome-wide studies of protein-protein interactions, most protein complexes have to be modeled by high-throughput computational approaches. The existing X-ray structures of protein complexes provide templates for ∼ 20% of all known interactions. Thus, docking of the remaining complexes has to be based on independently modeled structures of the monomers. Application of the docking methodologies, both template-based and template-free, to inherently inaccurate models has its limitations. Thus, docking benchmarking on a large systematic set of protein models at different levels of structural accuracy of the monomers is important. Currently available benchmark sets of protein-protein complexes (DOCKGROUND unbound set, Boston University Benchmark set, etc.) are limited to the X-ray structures in the bound and unbound forms and thus are unsuitable for such studies. We present a set of models built for 99 binary complexes from the DOCKGROUND unbound set. For each monomer in the dataset, six models were generated with Ca RMSD between the native and the modeled structures in 1 - 6 A range. The models were built by a combination of single-template homology modeling and Nudget Elastic Band (NEB) methodology. The dataset will be incorporated into the DOCKGROUND public resource (dockground.bioinformatics.ku.edu).