Leveraging GPU-Accelerated Molecular Dynamics Simulations to Compute and Analyze the 4D Chemical Descriptor Space of ERK2 Kinase Inhibitors
Jeremy Ash (1) and Denis Fourches (1,2)
1 Bioinformatics Research Center, North Carolina State University, Raleigh, USA.
2 Department of Chemistry, North Carolina State University, Raleigh, USA.
Quantitative Structure-Activity Relationship (QSAR) models typically rely on 2D and 3D molecular descriptors to characterize chemicals and forecast their experimental activities. When the biological target is known, structure-based molecular docking can also be used. In a previous study , we showed that even the most reliable QSAR models and docking techniques were not capable of accurately ranking a set of inhibitors for the ERK2 kinase, a key player in various types of cancer. Herein, building upon early-on 4D-QSAR studies , we calculated and analyzed a series of 4D chemical descriptors computed from the molecular dynamics (MD) trajectories of each ERK2-ligand complex. First, the docking of 122 ERK2 ligands with known binding affinities was accomplished using Schrodinger’s Glide software, whereas solvent-explicit MD simulations were performed using the GPU-accelerated Desmond program. These simulations (20 ns, NVT, 300K, TIP3P, 1fs) were performed using a GPU workstation based on dual Xeon processors and a 4-way Nvidia TitanX setup, allowing us to simultaneously perform multiple simulations on the workstation. Second, we calculated each 4D descriptor as a weighted distribution of a 3D WHIM descriptor computed for representative samples of the ligand’s conformations over the MD simulation. Third, we analyzed the dataset of 122 inhibitors in the 4D chemical descriptor space. Our results showed that 4D descriptors were able to distinguish the most active ERK2 inhibitors from the moderate/weak actives and inactives. Our presentation will focus on the analysis and visualization of the 4D chemical space. This study represents the largest attempt of utilizing MD-extracted descriptors to model a series of kinase inhibitors. These descriptors will enable the next generation of MD-QSAR models for computer-aided lead optimization and analogue prioritization. Importantly, this project could not have been possible without the GPU acceleration for our 122 independent, solvent-explicit MD simulations.
 Fourches D, Muratov E, Ding F, Dokholyan NV, Tropsha A. Predicting binding affinity of CSAR ligands using both structure-based and ligand-based approaches. J Chem Inf Model. 2013, 53, 1915-1922.
 Andrade CH, Pasqualoto KF, Ferreira EI, Hopfinger AJ. 4D-QSAR: perspectives in drug design. Molecules. 2010, 15, 3281-3294.