arXiv subjects

Oliver Beckstein

Publications and source records attributed to Oliver Beckstein.

10 recordsLinked to original sources

arXiv2026

Streaming Molecular Dynamics Simulation Data for On-the-fly Processing and Analysis

Only a small fraction of the data generated in state-of-the-art all-atom multi-microsecond molecular dynamics (MD) simulations is typically analyzed. With femtosecond integration steps, microsecond simulations generate billions of time steps containing atomic positions, velocities, and forces, often corresponding to petabytes of data. Since this exceeds practical storage capacities, only a subset of frames is usually written to trajectory files at intervals of 10-100 ps. While sufficient for ensemble averages and slow dynamics, this approach discards information on fast processes such as molecular vibrations, solvent dynamics, short-lived transition states, and transport phenomena that are directly related to macroscopic properties and experimental observables. Here, we introduce a streaming framework for MD simulations that provides direct access to all data generated during a running simulation. Instead of writing high-frequency output to disk, the framework enables user-defined analysis and processing routines to access simulation data in real time. To achieve this, we extend the Interactive Molecular Dynamics (IMD) protocol and implement an enhanced version, termed IMDv3, in the MD packages GROMACS, NAMD, and LAMMPS. We further introduce the Python package imdclient, which receives IMDv3 data streams and exposes them to external applications. To maximize usability, we add an imdclient-based reader to the MDAnalysis package, enabling streamed simulation data to be analyzed alongside conventional trajectory files. Benchmark results show that streaming can improve performance compared to simulations with high-frequency trajectory output. Example applications include live monitoring of custom variables, evaluation of velocity time-correlation functions with fast fluctuations, and real-time analysis of membrane pore currents.

cond-mat.stat-mech↗

arXiv2019

General Principles of Secondary Active Transporter Function

Transport of ions and small molecules across the cell membrane against electrochemical gradients is catalyzed by integral membrane proteins that use a source of free energy to drive the energetically uphill flux of the transported substrate. Secondary active transporters couple the spontaneous influx of a "driving" ion such as Na$^+$ or H$^+$ to the flux of the substrate. The thermodynamics of such cyclical non-equilibrium systems are well understood and recent work has focused on the molecular mechanism of secondary active transport. The fact that these transporters change their conformation between an inward-facing and outward-facing conformation in a cyclical fashion, called the alternating access model, is broadly recognized as the molecular framework in which to describe transporter function. High resolution structures and detailed computer simulations lead to the recognition of common molecular-level principles between disparate transporter families. Inverted repeat symmetry in secondary active transporters has shed light on how protein structures can encode a bi-stable two-state system. Three broad classes of alternating access transitions have been described as rocker-switch, rocking-bundle, and elevator mechanisms. Transporters can be understood as gated pores with at least two coupled gates that map to distinct parts of the transporter protein. Enumerating all distinct gate states naturally includes occluded states in the alternating access picture and suggests observable protein conformations. By connecting the possible conformational states and ion/substrate bound states in a kinetic model, a unified picture emerges in which symporter, antiporter, and uniporter function are extremes in a continuum of functionality. We briefly discuss how biological complexity may be integrated in quantitative kinetic models to provide a bridge from structure to function.

q-bio.BM↗

arXiv2019

Parallel Performance of Molecular Dynamics Trajectory Analysis

The performance of biomolecular molecular dynamics simulations has steadily increased on modern high performance computing resources but acceleration of the analysis of the output trajectories has lagged behind so that analyzing simulations is becoming a bottleneck. To close this gap, we studied the performance of parallel trajectory analysis with MPI and the Python MDAnalysis library on three different XSEDE supercomputers where trajectories were read from a Lustre parallel file system. Strong scaling performance was impeded by stragglers, MPI processes that were slower than the typical process. Stragglers were less prevalent for compute-bound workloads, thus pointing to file reading as a bottleneck for scaling. However, a more complicated picture emerged in which both the computation and the data ingestion exhibited close to ideal strong scaling behavior whereas stragglers were primarily caused by either large MPI communication costs or long times to open the single shared trajectory file. We improved overall strong scaling performance by either subfiling (splitting the trajectory into separate files) or MPI-IO with Parallel HDF5 trajectory files. The parallel HDF5 approach resulted in near ideal strong scaling on up to 384 cores (16 nodes), thus reducing trajectory analysis times by two orders of magnitude compared to the serial approach.

cs.DC↗

arXiv2019

Learning Everywhere: Pervasive Machine Learning for Effective High-Performance Computation

The convergence of HPC and data-intensive methodologies provide a promising approach to major performance improvements. This paper provides a general description of the interaction between traditional HPC and ML approaches and motivates the Learning Everywhere paradigm for HPC. We introduce the concept of effective performance that one can achieve by combining learning methodologies with simulation-based approaches, and distinguish between traditional performance as measured by benchmark scores. To support the promise of integrating HPC and learning methods, this paper examines specific examples and opportunities across a series of domains. It concludes with a series of open computer science and cyberinfrastructure questions and challenges that the Learning Everywhere paradigm presents.

cs.DC↗

arXiv2018

Task-parallel Analysis of Molecular Dynamics Trajectories

Different parallel frameworks for implementing data analysis applications have been proposed by the HPC and Big Data communities. In this paper, we investigate three task-parallel frameworks: Spark, Dask and RADICAL-Pilot with respect to their ability to support data analytics on HPC resources and compare them with MPI. We investigate the data analysis requirements of Molecular Dynamics (MD) simulations which are significant consumers of supercomputing cycles, producing immense amounts of data. A typical large-scale MD simulation of a physical system of O(100k) atoms over {\mu}secs can produce from O(10) GB to O(1000) GBs of data. We propose and evaluate different approaches for parallelization of a representative set of MD trajectory analysis algorithms, in particular the computation of path similarity and leaflet identification. We evaluate Spark, Dask and RADICAL-Pilot with respect to their abstractions and runtime engine capabilities to support these algorithms. We provide a conceptual basis for comparing and understanding different frameworks that enable users to select the optimal system for each application. We also provide a quantitative performance analysis of the different algorithms across the three frameworks.

cs.DC↗

arXiv2015

Path Similarity Analysis: a Method for Quantifying Macromolecular Pathways

Diverse classes of proteins function through large-scale conformational changes; sophisticated enhanced sampling methods have been proposed to generate these macromolecular transition paths. As such paths are curves in a high-dimensional space, they have been difficult to compare quantitatively, a prerequisite to, for instance, assess the quality of different sampling algorithms. The Path Similarity Analysis (PSA) approach alleviates these difficulties by utilizing the full information in 3N-dimensional trajectories in configuration space. PSA employs the Hausdorff or Fréchet path metrics---adopted from computational geometry---enabling us to quantify path (dis)similarity, while the new concept of a Hausdorff-pair map permits the extraction of atomic-scale determinants responsible for path differences. Combined with clustering techniques, PSA facilitates the comparison of many paths, including collections of transition ensembles. We use the closed-to-open transition of the enzyme adenylate kinase (AdK)---a commonly used testbed for the assessment enhanced sampling algorithms---to examine multiple microsecond equilibrium molecular dynamics (MD) transitions of AdK in its substrate-free form alongside transition ensembles from the MD-based dynamic importance sampling (DIMS-MD) and targeted MD (TMD) methods, and a geometrical targeting algorithm (FRODA). A Hausdorff pairs analysis of these ensembles revealed, for instance, that differences in DIMS-MD and FRODA paths were mediated by a set of conserved salt bridges whose charge-charge interactions are fully modeled in DIMS-MD but not in FRODA. We also demonstrate how existing trajectory analysis methods relying on pre-defined collective variables, such as native contacts or geometric quantities, can be used synergistically with PSA, as well as the application of PSA to more complex systems such as membrane transporter proteins.

q-bio.QM↗

arXiv2009

Opening a hydrophobic gate: the nicotinic acetylcholine receptor as an example

To what extent must a hydrophobic gate expand for the channel to count as open? We address this question using the nicotinic acetylcholine receptor (nAChR) as the exemplar. The nAChR is an integral membrane protein which forms a cation selective channel gated by neurotransmitter binding to its extracellular domain. A hydrophobic gating model has been proposed for the nAChR, whereby the pore is incompletely occluded in the closed state channel, with a narrow hydrophobic central gate region which presents an energetic barrier to ion permeation. The nAChR pore is lined by a parallel bundle of five M2 alpha-helices, with the gate formed by three rings of hydrophobic sidechains (9', 13', and 17' of M2). A number of models have been proposed to describe the nature of the conformational change underlying the closed to open transition of the nAChR. These models involve different degrees of M2 helix displacement, rotation, and/or kinking. In this study, we use a simple pore expansion method (previously used to model opening of potassium channels) to generate a series of progressively wider models of the nAChR transmembrane domain. Continuum electrostatics calculations are used to assess the change in the barrier height of the hydrophobic gate as a function of pore expansion. The results suggest that an increase in radius of Delta r ~ 1.5 angstrom is sufficient to functionally open the pore without, for example, a requirement for rotation of the M2 helices. This is evaluated in the context of current mutational and structural data on the nAChR and its homologues.

q-bio.BM↗

arXiv2005

A Hydrophobic Gate in an Ion Channel: The Closed State of the Nicotinic Acetylcholine Receptor

The nicotinic acetylcholine receptor (nAChR) is the prototypic member of the `Cys-loop' superfamily of ligand-gated ion channels which mediate synaptic neurotransmission, and whose other members include receptors for glycine, gamma-aminobutyric acid, and serotonin. Cryo-electron microscopy has yielded a three dimensional structure of the nAChR in its closed state. However, the exact nature and location of the channel gate remains uncertain. Although the transmembrane pore is constricted close to its center, it is not completely occluded. Rather, the pore has a central hydrophobic zone of radius about 3 A. Model calculations suggest that such a constriction may form a hydrophobic gate, preventing movement of ions through a channel. We present a detailed and quantitative simulation study of the hydrophobic gating model of the nicotinic receptor, in order to fully evaluate this hypothesis. We demonstrate that the hydrophobic constriction of the nAChR pore indeed forms a closed gate. Potential of mean force (PMF) calculations reveal that the constriction presents a barrier of height ca. 10 kT to the permeation of sodium ions, placing an upper bound on the closed channel conductance of 0.3 pS. Thus, a 3 A radius hydrophobic pore can form a functional barrier to the permeation of a 1 A radius Na+ ion. Using a united atom force field for the protein instead of an all atom one retains the qualitative features but results in differing conductances, showing that the PMF is sensitive to the detailed molecular interactions.

q-bio.BM↗

arXiv2004

The influence of geometry, surface character and flexibility on the permeation of ions and water through biological pores

A hydrophobic constriction site can act as an efficient barrier to ion and water permeation if its diameter is less than the diameter of an ion's first hydration shell. This hydrophobic gating mechanism is thought to operate in a number of ion channels, e.g. the nicotinic receptor, bacterial mechanosensitive channels (MscL and MscS) and perhaps in some potassium channels (e.g. KcsA, MthK, and KvAP). Simplified pore models allow one to investigate the primary characteristics of a conduction pathway, namely its geometry (shape, pore length, and radius), the chemical character of the pore wall surface, and its local flexibility and surface roughness. Our extended (ca. 0.1 μs) molecular dynamic simulations show that a short hydrophobic pore is closed to water for radii smaller than 0.45 nm. By increasing the polarity of the pore wall (and thus reducing its hydrophobicity) the transition radius can be decreased until for hydrophilic pores liquid water is stable down to a radius comparable to a water molecule's radius. Ions behave similarly but the transition from conducting to non-conducting pores is even steeper and occurs at a radius of 0.65 nm for hydrophobic pores. The presence of water vapour in a constriction zone indicates a barrier for ion permeation. A thermodynamic model can explain the behaviour of water in nanopores in terms of the surface tensions, which leads to a simple measure of "hydrophobicity" in this context. Furthermore, increased local flexibility decreases the permeability of polar species. An increase in temperature has the same effect, and we hypothesise that both effects can be explained by a decrease in the effective solvent-surface attraction which in turn leads to an increase in the solvent-wall surface free energy.

q-bio.SC↗

arXiv2003

Liquid-vapor oscillations of water in hydrophobic nanopores

Water plays a key role in biological membrane transport. In ion channels and water-conducting pores (aquaporins), one dimensional confinement in conjunction with strong surface effects changes the physical behavior of water. In molecular dynamics simulations of water in short (0.8 nm) hydrophobic pores the water density in the pore fluctuates on a nanosecond time scale. In long simulations (460 ns in total) at pore radii ranging from 0.35 nm to 1.0 nm we quantify the kinetics of oscillations between a liquid-filled and a vapor-filled pore. This behavior can be explained as capillary evaporation alternating with capillary condensation, driven by pressure fluctuations in the water outside the pore. The free energy difference between the two states depends linearly on the radius. The free energy landscape shows how a metastable liquid state gradually develops with increasing radius. For radii larger than ca. 0.55 nm it becomes the globally stable state and the vapor state vanishes. One dimensional confinement affects the dynamic behavior of the water molecules and increases the self diffusion by a factor of two to three compared to bulk water. Permeabilities for the narrow pores are of the same order of magnitude as for biological water pores. Water flow is not continuous but occurs in bursts. Our results suggest that simulations aimed at collective phenomena such as hydrophobic effects may require simulation times longer than 50 ns. For water in confined geometries, it is not possible to extrapolate from bulk or short time behavior to longer time scales.

physics.bio-ph↗