An integrated approach to protein-protein docking

Rong Chen,Z. Weng

2004-01-01

Abstract:Structural genomics aims to determine the 3-dimensional (3D) structure of every protein in the cell, and high throughput techniques such as yeast-II-hybrid and protein microarrays will eventually identify nearly all protein-protein interactions. Considerable attention is paid currently to the computational methods of investigating these interactions. One such method is protein-protein docking, which predicts the 3D structure of a protein complex starting from the 3D structures of two interacting partners. This dissertation describes an integrated approach to protein-protein docking: an initial-stage docking program ZDOCK followed by a refinement-stage docking program RDOCK. ZDOCK treats the receptor and the ligand as rigid bodies and uses conformation-change-tolerant scoring functions to fully explore the 3 rotational and 3 translational degrees of freedom. The rotational space is searched explicitly and the translational space is searched using a Fast Fourier Transform (FFT) algorithm to speed up the process. Several scoring functions are investigated, including a novel pairwise shape complementarity (PSC) function, Atomic Contact Energy (ACE)-based desolvation energy, Interface Atomic Contact Energy (IFACE), and coulombic electrostatic energy. The top 2000 predictions from MOCK are then minimized in RDOCK using a three-stage energy-minimization scheme with a special treatment of ionic side chains. The minimized structures are then re-evaluated using a more sophisticated binding free energy, which combines the molecular mechanics software CHARMM with a statistical desolvation energy ACE. To evaluate the performance of my software as well as other docking groups' software, a non-redundant docking benchmark has been built by collecting all protein complexes with both complex structure and at least one independently determined monomer structure available. When tested against the benchmark, my integrated approach identified a near-native structure (within 2.5A from the experimental structure) as the top prediction for 37% of test cases, and within the top 4 predictions for 49% of test cases. Its superior performance has also been demonstrated in a community-wide protein docking blind test, CAPRI, where we successfully predicted the complex structures for 4 targets and had some success in 3 others. An automatic docking server has also been built to serve the docking community.

Engineering