Accurate Prediction of Complex Structure and Affinity for a Flexible Protein Receptor and Its Inhibitor

Abstract

In order to predict the accurate binding configuration as well as the binding affinity for a flexible protein receptor and its inhibitor drug, enhanced sampling with multicanonical molecular dynamics (McMD) simulation and thermodynamic integration (TI) were combined as a general drug docking method. CDK2, cyclin-dependent kinase 2, is involved in the cell cycle regulation. Malfunctions in CDK2 can cause tumorigenesis, and thus it is a potential drug target. Here, we performed a long McMD simulation for docking the inhibitor CS3 to CDK2 starting from the unbound structure. Subsequently, a potential binding/unbinding pathway was given from the multicanonical ensemble, and the binding free energy was readily computed by TI along the pathway. Using this combination, the correct binding configuration of CS3 to CDK2 was obtained, and its affinity coincided well with the experimental value.

Exhaustive search of the configurational space of heat‐shock protein 90 with its inhibitor by multicanonical molecular dynamics based dynamic docking

“…Recently, dynamic docking to explore binding configurations between receptor proteins and their ligands by molecular dynamics (MD) simulations [ 1–5 ] has become possible by the use of GPGPU‐accelerated clusters [ 6 ] , which are clusters that make use of General Purpose Graphics Processing Units, and MD‐optimized supercomputers such as MDGRAPE. [ 7 ] Unlike traditional docking, dynamic docking enables accurate and extensive sampling of phase space by MD simulations, treating both ligand and protein receptor as flexible, while including solvation and entropic effects.…”

“…We have previously applied McMD to the conformational sampling of proteins and peptides, [ 17,18 ] loop structure prediction of an antibody, [ 19 ] and for the docking simulations between receptors and their ligands. [ 1,2,4,5 ] Our recent McMD docking studies restrained the ligand inside a cylindrical region covering both the binding site and the bulk region, enabling us to successfully predict the native binding configuration in addition to generating a smooth binding/unbinding path. [ 2,4,5 ] However, to accurately predict the accurate ligand binding configuration without prior knowledge of the approximate location of the binding site by molecular docking still remains challenging, because a considerably more exhaustive search covering the entire receptor protein would be required.…”

“…[ 33 ] To further assess the strength of our dynamic docking method and determine whether it could predict the native binding configuration without prior knowledge of the binding pocket, we sought to perform the simulations without any cylindrical restraints on the ligand. Previously, we performed simulations using a cylinder on cyclin‐dependent kinase 2 (CDK2) with one of its inhibitors, [ 2 ] BACE with one of its inhibitors, [ 4 ] and an antibody and its antigen, an amyloid‐β (Aβ) peptide. [ 5 ] Given the compact and globular shape of Hsp90, with it consisting of a central binding pocket with otherwise few discernable pockets on its surface, makes it a suitable benchmark candidate.…”

Improving the Accuracy of Protein‐Ligand Binding Mode Prediction Using a Molecular Dynamics‐Based Pocket Generation Approach

“…Additionally, an MD study of several tens of proteins suggested that pocket conformations in the ligand‐bound state could not be fully produced by a ~100 ns simulation of the apo‐protein itself . To circumvent these limitations, several advanced MD methods that greatly enhance conformational sampling have been developed . Although these methods overcome both of the above‐mentioned problems, they have a high computational cost (e.g., MD simulations for a total of more than 10 μs, which cost several thousand CPU days, are required for evaluating a single protein‐compound pair).…”

“…[11] To circumvent these limitations, several advanced MD methods that greatly enhance conformational sampling have been developed. [12][13][14] Although these methods overcome both of the above-mentioned problems, they have a high computational cost (e.g., MD simulations for a total of more than 10 μs, which cost several thousand CPU days, are required for evaluating a single protein-compound pair).…”