The NVIDIA GPU Award for Best GPU Poster Finalists for Denver (spring 2015)

Post date: Jan 28, 2015 7:39:09 PM

The COMP Division is excited to announce the NVIDIA GPU Award finalists for the Best GPU Poster at the Denver ACS meeting (spring 2015). Please visit the COMP award winners and the other excellent COMP posters at the COMP Poster Session on Tuesday, March 24, 2015 from 6pm to 8pm in Hall B2 of the Colorado Convention Center. More information about the NVIDIA GPU Award for Best GPU Poster can be found here.

Ligand-specific conformational changes in CCR7 coupled to signaling pathway selection

David Lo, Zied Gaieb, Dimitrios Morikis.University of California, Riverside

Chemokine receptor type 7 (CCR7) belongs to family one of the G-Protein Coupled Receptor (GPCR) super-family. It is activated by CCL19 and CCL21, which are expressed in different parts of the body as a gradient to aid in homing of T cells and antigen-presenting dendritic cells to the lymph nodes. Although both ligands have similar structures and activate the same receptor (CCR7), they induce distinct signaling pathways. While both ligands mediate their signaling through G-protein and GRK6, only CCL19 induces CCR7 desensitization and internalization through phosphorylation by GRK3 and recruitment of β-arrestin. The functional diversity of receptor-ligand binding is related to decoupled or partially decoupled conformational changes. Different conformational changes induced by biased ligands are still not fully understood. We aim to delineate the structural elements that are responsible for biased receptor activation, and the conformational changes and long-range motions that link extracellular ligand binding with intracellular protein binding and selection of signaling pathways. Using GPU-accelerated molecular dynamics (MD) software, NAMD, we performed series of MD simulations of the free receptor and CCL21-CCR7 complex to determine long-range conformational changes associated with receptor activation pathways upon ligand binding. Differences in the conformational changes between both systems were quantitatively assessed using a range of MD analysis protocols, including principle component analysis (PCA) and dynamic cross-correlation analysis (DCC). Comparing the correlation maps, we saw that the presence of the ligand introduced collective motions within larger CCR7 domains. The intracellular region of trans-membrane (TM) 5 and 6 and the intracellular loop 3 (ICL3) are highly correlated protein regions. This finding may be necessary to initiate the large opening between the intracellular regions of TM5 and TM6 that has been shown necessary to accommodate G-protein binding in the β2-adrenergic receptor. The average conformations of the free and bound receptor reveal differences in the behavior of ICL2. In this study, we were able to distinguish the different patterns of ligand-induced correlated motions, important for activation of the receptor. Examination of these different motions allowed us to isolate important receptor motifs needed to guide the recognition and initiation of ligand-induced conformational changes.

Theoretical view of the C3d-CR2 binding controversy

Rohith Mohan, Ronald Gorham, Dimitrios Morikis. University of California, Riverside

The complement system, consisting of several plasma proteins, is an important part of the innate immune system. The interaction of complement fragment C3d and complement receptor 2 (CR2) plays a crucial role as a link between innate and adaptive immunity, leading to enhancement of B cell mediated antibody production during initial complement response to infection. Over the past decade, there has been much debate over the binding mode of this interaction. An initial cocrystal structure (PDB: 1GHQ) was published in 2001, in which the only interactions observed were between the SCR2 domain of CR2 and a side-face of C3d whereas a cocrystal structure (PDB: 3OED) published in 2011 showed both the SCR1 and SCR2 domains of CR2 interacting with an acidic patch on the concave surface of C3d. The initial 1GHQ structure is at odds with the majority of existing biochemical data and the publication of the 3OED structure renewed uncertainty regarding the physiological relevance of 1GHQ, suggesting that crystallization may have been influenced by the presence of zinc acetate in the crystallization process. In our study, we used a variety of computational approaches such as explicit-solvent and steered molecular dynamics simulations using the GPU-accelerated software, NAMD, to gain insight into the binding mode between C3d and CR2. We demonstrate that the binding site at the acidic patch (3OED) is electrostatically more favorable, exhibits better structural and dissociative stability, specifically at the SCR1 domain, and has higher binding affinity than the 1GHQ binding mode. We also observe that nonphysiological zinc ions enhance the formation of the C3d-CR2 complex at the side face of C3d (1GHQ) through increases in electrostatic favorability, intermolecular interactions, dissociative character and overall energetic favorability. These results provide a theoretical basis for the association of C3d-CR2 at the acidic cavity of C3d and provide an explanation for binding of CR2 at the side face of C3d in the presence of non physiological zinc ions.

Entropically driven CK-ASB9 interaction: How GPU-enabled computing provided unique insight into an intrinsically disordered 116 kDa protein complex

Elizabeth Komives, Jamie Schiffer, Jonathan Parnell, Rommie Amaro. University of California, San Diego

The ankyrin-repeat and SOCS box protein 9 (ASB9) targets creatine kinases (CK) for degradation, a process crucial to the regulation of ATP/ADP levels at the cell surface of cardiomyocytes. It has been show that the elevation of CK levels in failing mouse hearts increases survival rates threefold. Elevating CK levels in cardiomyocytes in vivo could be achieved by disruption of the CK-ASB9 complex with a small molecule. However, due to the size and structural mobility of intrinsically disordered protein (IDP) complexes like CK-ASB9, X-ray crystallography and NMR spectroscopy, alone, cannot adequately describe such structural ensembles. In this work, multiple lines of experimental data are integrated with computational docking and GPU-enabled simulations to produce a highly accurate model of a large IDP complex. All-atom, explicit-solvent, molecular dynamics (MD) simulations produced a 1.3 μs ensemble of the CK-ASB9 complex, performed on GPUs using the CUDA version of PMEMD in AMBER. From these simulations, a large subset of the total ensemble correlated well with experimental SAXS profiles (χ2 = 0.939) and Rgs, HDX-MS protection factors, ITC enthalpies based on MMGBSA calculations, and mutation data. Where conventional structural biology approaches fail to provide adequate results, high-throughput MD simulations can produce atomistic level trajectories of IDP complexes that are long enough to adequately sample IDP's unique conformational landscape and allow researchers to identify new potential avenues for drug discovery. Such high-throughput simulations are only feasible owing to the enormous speedups provided by GPU-enabled computing. In the image to the right, the structural ensemble of CK(teal)-ASB923-252(orange) shows the disordered ADP/ATP binding loops (yellow) and HDX protected residues (red).

GPU-enabled real-time electron dynamics of large light-harvesting systems in explicit solvent

Bryan Wong, Ma. Belen Oviedo. University of California, Riverside

image of aligned and overlaid 3D protein structures

Photo-initiated charge-transfer processes play a central role in biophysical systems such as human vision and photosynthesis. While researchers have successfully mimicked these processes for simple isolated systems, our understanding of photo-initiated mechanisms in realistic and complex environments is still in its infancy. In particular, recent experiments have shown that simple descriptions of solvent interactions (either via classical force fields or effective solvent models) are unable to accurately capture the electron dynamics in these environments. These ongoing observations open an entirely new computational area of research in the properties of light-activated processes in explicit solvent, with the opportunity to deeply understand the real-time electron dynamics in large complex systems.

To this end, we have developed a new real-time time-dependent density functional tight binding (RT-TDDFTB) code that efficiently runs on massively-parallelized GPUs. This GPU-enhanced capability allows efficient calculations of real-time electron dynamics of donor-acceptor complexes in the presence of explicit solvent molecules - all treated at the quantum mechanical level. Furthermore, and most importantly, the use of GPUs allows us to calculate the electron dynamics of large solvated systems (~10,000 atoms), whereas conventional approaches are computationally limited to only hundreds of atoms. Using a custom-built GPU cluster with a GTX Titan processor in our laboratory, we are able to understand and rationalize electron-hole recombination effects as a function of solvent polarity, configuration, and energy transfer. Furthermore, this new computational capability gives us mechanistic insight into the electron dynamics of complex environments with the goal of probing charge-transfer dynamics in large systems.

GPU-accelerated implementations of advanced electronic structure methods in Q-Chem

Anna Krylov, Kaushik Nanda, Evgeny Epifanovsky. University of Southern California

Electronic structure methods based on the principles of quantum chemistry play an important role in aiding, designing and explaining experiments in chemistry and spectroscopy. Electronic structure methods such as coupled-cluster (CC) and equation-of-motion coupled-cluster (EOM-CC) are robust techniques for dealing with the ground-state and excited states of molecules with high accuracy albeit with a high computational cost. Faster and more efficient implementations of these methods are desirable in many applications. We will present new GPU-accelerated implementations for the CC and EOM-CC methods in the ab initio quantum chemistry package Q-Chem. Performance and scaling analysis of these CUDA- based implementations against the corresponding CPU implementations will be demonstrated.