Showing papers by "David E. Shaw published in 2014"

Anton 2: raising the bar for performance and programmability in a special-purpose molecular dynamics supercomputer

Abstract: Anton 2 is a second-generation special-purpose supercomputer for molecular dynamics simulations that achieves significant gains in performance, programmability, and capacity compared to its predecessor, Anton 1. The architecture of Anton 2 is tailored for fine-grained event-driven operation, which improves performance by increasing the overlap of computation with communication, and also allows a wider range of algorithms to run efficiently, enabling many new software-based optimizations. A 512-node Anton 2 machine, currently in operation, is up to ten times faster than Anton 1 with the same number of nodes, greatly expanding the reach of all-atom bio molecular simulations. Anton 2 is the first platform to achieve simulation rates of multiple microseconds of physical time per day for systems with millions of atoms. Demonstrating strong scaling, the machine simulates a standard 23,558-atom benchmark system at a rate of 85 µs/day -- 180 times faster than any commodity hardware platform or general-purpose supercomputer.

How IGF-1 activates its receptor

Abstract: Hormones are chemicals that are produced to carry signals around the body. Mammals, including humans, need hormones called insulin and insulin-like growth factors (or IGF for short) to grow and develop normally. These hormones bind to, and activate, specific proteins—known as receptors—that span from the outside of the cell to the inside through the cell's surface membrane. The insulin and IGF receptors are complex molecules, each composed of two identical protein subunits that are linked together. The hormones bind to the part of the proteins (known as the extracellular domains) that are on the outside surface of cells. When no hormone is bound to the receptor, these two extracellular domains form an inverted ‘V’ shape and the two regions that cross the membrane are held far apart. It was unclear, however, how this shape changes when the hormone binds, and how this change in shape activates the receptor. Kavran et al. have now compared the known three-dimensional structures of the extracellular domains of the insulin receptor, both with and without a molecule of insulin bound to it. This comparison highlighted an interaction between the two receptor subunits that was disrupted when insulin was bound to the receptor. This interaction appeared to stabilize the inverted ‘V’ structure and hold apart the parts of the proteins that span the surface—which in turn separates the regions of the proteins that are inside the cell. Kavran et al. suggest that this separation keeps the insulin receptor in an inactive state. Removing either the whole extracellular domain of the IGF receptor—or a smaller portion that keeps the regions that cross the cell membrane separate—resulted in the receptor being activated, even when insulin-like growth factor was not bound to the receptor. Kavran et al. then confirmed that when a molecule of insulin-like growth factor binds to the extracellular domain of the IGF receptor, it causes a shape change that moves the parts of the receptor that span the cell membrane closer together. This results in the regions of the receptor proteins located inside the cell adding chemical tags, called phosphate groups, to one another—which activates the receptor. When there are problems with how these receptors are activated or inactivated in humans, serious disorders including diabetes and cancer can occur. As such, the findings of Kavran et al. might help future work aimed at regulating the activation of the insulin and IGF receptors to treat these and other diseases.

Molecular basis for pseudokinase-dependent autoinhibition of JAK2 tyrosine kinase

Abstract: Janus kinase-2 (JAK2) mediates signaling by various cytokines, including erythropoietin and growth hormone. JAK2 possesses tandem pseudokinase and tyrosine-kinase domains. Mutations in the pseudokinase domain are causally linked to myeloproliferative neoplasms (MPNs) in humans. The structure of the JAK2 tandem kinase domains is unknown, and therefore the molecular bases for pseudokinase-mediated autoinhibition and pathogenic activation remain obscure. Using molecular dynamics simulations of protein-protein docking, we produced a structural model for the autoinhibitory interaction between the JAK2 pseudokinase and kinase domains. A striking feature of our model, which is supported by mutagenesis experiments, is that nearly all of the disease mutations map to the domain interface. The simulations indicate that the kinase domain is stabilized in an inactive state by the pseudokinase domain, and they offer a molecular rationale for the hyperactivity of V617F, the predominant JAK2 MPN mutation.

Structural analysis of the EGFR/HER3 heterodimer reveals the molecular basis for activating HER3 mutations

Abstract: The human epidermal growth factor receptor (HER) tyrosine kinases homo- and heterodimerize to activate downstream signaling pathways. HER3 is a catalytically impaired member of the HER family that contributes to the development of several human malignancies and is mutated in a subset of cancers. HER3 signaling depends on heterodimerization with a catalytically active partner, in particular epidermal growth factor receptor (EGFR) (the founding family member, also known as HER1) or HER2. The activity of homodimeric complexes of catalytically active HER family members depends on allosteric activation between the two kinase domains. To determine the structural basis for HER3 signaling through heterodimerization with a catalytically active HER family member, we solved the crystal structure of the heterodimeric complex formed by the isolated kinase domains of EGFR and HER3. The structure showed HER3 as an allosteric activator of EGFR and revealed a conserved role of the allosteric mechanism in activation of HER family members through heterodimerization. To understand the effects of cancer-associated HER3 mutations at the molecular level, we solved the structures of two kinase domains of HER3 mutants, each in a heterodimeric complex with the kinase domain of EGFR. These structures, combined with biochemical analysis and molecular dynamics simulations, indicated that the cancer-associated HER3 mutations enhanced the allosteric activator function of HER3 by redesigning local interactions at the dimerization interface.

Membrane interaction of bound ligands contributes to the negative binding cooperativity of the EGF receptor.

Abstract: The epidermal growth factor receptor (EGFR) plays a key role in regulating cell proliferation, migration, and differentiation, and aberrant EGFR signaling is implicated in a variety of cancers. EGFR signaling is triggered by extracellular ligand binding, which promotes EGFR dimerization and activation. Ligand-binding measurements are consistent with a negatively cooperative model in which the ligand-binding affinity at either binding site in an EGFR dimer is weaker when the other site is occupied by a ligand. This cooperativity is widely believed to be central to the effects of ligand concentration on EGFR-mediated intracellular signaling. Although the extracellular portion of the human EGFR dimer has been resolved crystallographically, the crystal structures do not reveal the structural origin of this negative cooperativity, which has remained unclear. Here we report the results of molecular dynamics simulations suggesting that asymmetrical interactions of the two binding sites with the membrane may be responsible (perhaps along with other factors) for this negative cooperativity. In particular, in our simulations the extracellular domains of an EGFR dimer spontaneously lay down on the membrane in an orientation in which favorable membrane contacts were made with one of the bound ligands, but could not be made with the other. Similar interactions were observed when EGFR was glycosylated, as it is in vivo.

Assessing the Accuracy of Two Enhanced Sampling Methods Using EGFR Kinase Transition Pathways: The Influence of Collective Variable Choice

Abstract: Structurally elucidating transition pathways between protein conformations gives deep mechanistic insight into protein behavior but is typically difficult. Unbiased molecular dynamics (MD) simulations provide one solution, but their computational expense is often prohibitive, motivating the development of enhanced sampling methods that accelerate conformational changes in a given direction, embodied in a collective variable. The accuracy of such methods is unclear for complex protein transitions, because obtaining unbiased MD data for comparison is difficult. Here, we use long-time scale, unbiased MD simulations of epidermal growth factor receptor kinase deactivation as a complex biological test case for two widely used methods—steered molecular dynamics (SMD) and the string method. We found that common collective variable choices, based on the root-mean-square deviation (RMSD) of the entire protein, prevented the methods from producing accurate paths, even in SMD simulations on the time scale of the unbi...

Unifying on-chip and inter-node switching within the Anton 2 network

Abstract: The design of network architectures has become increasingly complex as the chips connected by inter-node networkshave emerged as distributed systems in their own right, complete with their own on-chip networks. In Anton 2, a massively parallel special-purpose supercomputer for molecular dynamics simulations, we managed this complexity by reusing the on-chip network as a switch for inter-node traffic. This unified network approach introduces several design challenges. Maintaining fairness within the inter-node network is difficult, as each hop becomes a sequence of many on-chip routing decisions. We addressed this problem with an inverse-weighted arbiter that ensures fairness with low implementation costs. Balancing the load of inter-node traffic across the on-chip network is also critical, and we adopted an optimization approach to design an appropriate routing algorithm. Finally, the on-chip routers carry inter-node traffic, so they must implement inter-node virtual channels to avoid deadlock. In order to keep the routers small and fast, we developed a deadlock-free routing algorithm that reduces the number of virtual channels by one-third relative to previous approaches. The resulting Anton 2 network implementation efficiently utilizes its inter-node channels and provides low messaging latency, while occupying a modest amount of silicon area

Desmond/GPU Performance as of November 2014

Abstract: Desmond/GPU is a code, written in CUDA C++, that is designed for the execution of molecular dynamics (MD) simulations of biological systems on NVIDIA Graphics Processing Units (GPUs). This paper reports the performance that Desmond/GPU, running on a range of different GPUs and GPU cluster configurations, achieves on three biological system benchmarks as of November 2014. Our benchmark results show that on a single GPU, Desmond can deliver the same simulation throughput that it delivers on a dozen CPUs.