Proteins are both significant therapeutic agents and therapeutic targets. Structural studies of protein interactions are essential for the development of novel therapeutics with high binding affinities and specificities. Protein interactions often involve vital contributions from environmental conditions and interactions with non-peptide biomolecules. Thus, despite recent advances in protein-protein complex prediction algorithms, the diversity of biomolecular interactions requires a significant expansion of the computational toolbox. The recent rounds 20-27 of the Critical Assessment of PRotein Interactions (CAPRI), a community-wide blind structure prediction challenge, provided a testing platform for computational methods designed to address a wide range of challenges. We used RosettaDock and other novel Rosetta protocols to successfully predict one-third of the blind targets. For example, for DNase domain of Colicin E2 � Im2 immunity protein, we used RosettaDock and RosettaLigand with a TIP3P water model to predict the positions of water molecules at the interface, recovering 82% of the native contacts. For alpha-repeat Rep4-Rep2 and g-type lysozyme-PliG inhibitor complexes, we built homology models and used standard and pH-sensitive docking algorithms to successfully generate structures with interface RMSD values of 3.3 � and 2.0 �, respectively. We also developed a novel flexible sugar-protein docking protocol for structure prediction of the BT4661-heparin complex (results pending). Challenges remain in the generation of accurate homology models for protein mutants and sampling during global docking. For example, on proteins designed to bind influenza hemagglutinin, we identified only about half of the mutations that affect binding (T55: 54%; T56: 48%). The prediction of the structure of xylanase complex involving homology modeling and multi-domain docking pushed the limits of global conformational sampling and did not result in any successful prediction. The diversity of problems at hand requires computational algorithms to be versatile; our recent additions to the Rosetta suite expand the capabilities to encompass more realistic docking problems.

Blind predictions of structures and binding affinities of protein complexes

245th National Meeting (2013)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.