Self-Help

A VR framework for interacting with molecular simulations

Description
A VR framework for interacting with molecular simulations
Categories
Published
of 4
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
  A VR Framework for Interacting with Molecular Simulations Nicolas F´erey ∗ LBT - CNRS UPR 9080Olivier Delalande † LBT - CNRS UPR 9080Marc Baaden ‡ LBT - CNRS UPR 9080Gilles Grasseau § IDRIS - CNRS Abstract Molecular Dynamics is nowadays routinely used to complementexperimental studies and overcome some of their limitations. Inparticular, current experimental techniques do not allow to directlyobserve the full dynamics of a macromolecule at atomic detail.Molecular simulation engines provide time-dependent atomic posi-tions, velocities and system energies according to biophysical mod-els. Many molecular simulation engines can now compute a molec-ular dynamics trajectory of interesting biological systems in inter-activetime. Thisprogresshasleadtoanewapproachcalledinterac-tive molecular dynamics. It allows to control and visualise a molec-ular simulation in progress. We have developed a generic library,called  MDDriver  , in order to facilitate the implementation of suchinteractive simulations. It allows to easily create a network con-nection between a molecular user interface and a physically-basedsimulation. We use this library in order to study a real biomolecularsystem, simulatedbyvariousinteraction-enabledmolecularenginesand models. We use a classical molecular visualisation tool anda haptic device to control the dynamic behavior of the molecule.This approach provides encouraging results for interacting with abiomolecule and understanding its dynamics. Starting from this ini-tial success, we decided to use VR functionalities more intensively,by designing a VR framework dedicated to immersive and interac-tive molecular simulations. This framework is based on  MDDriver  ,on the visualisation toolkit  VTK  , and on the  vtkVRPN   library, whichencapsulates the  VRPN   library into  VTK  . CR Categories:  I.6.6 [Computing Methodologies]: Simulationand Modelling—Simulation Output Analysis H.5.2 [InformationSystems]: Information Interfaces and Presentation—Graphic UserInterface and Haptic I/O Keywords:  Scientific Visualisation, Interactive Molecular Dy-namics, Haptic Feedback, VTK, VRPN Introduction Molecular Dynamics (MD) allows researchers to obtain comple-mentary data with respect to experimental studies and to overcomesome of their limitations. Current experimental techniques do notallow to observe the full dynamics of a macromolecule at atomicdetail. In return, experiments provide the structures, i.e. the spatial ∗ e-mail: nicolas.ferey@ibpc.fr † e-mail: olivier.delalande@ibpc.fr ‡ e-mail:marc.baaden@ibpc.fr § e-mail:gilles.grasseau@idris.fr atomic positions, for numerous biomolecular systems, which areoften used as starting point for simulation studies.In order to predict, to explain and to understand experimental re-sults, researchers have developed a variety of biomolecular rep-resentations and algorithms. They allow to simulate the dynamicbehavior of macromolecules at different scales, ranging from de-tailed models using quantum mechanics or classical molecular me-chanics to more approximate representations [Baaden and Lavery2007]. These simulations are controlled  a priori  by complex andempirical settings. Most researchers visualise the result of theirsimulation once the computation is finished. Such post-simulationanalysis makes use of specific molecular user interfaces, by read-ing and visualising the molecular 3D configuration at each step of the simulation. This approach makes it difficult to interact with asimulation in progress. When a problem occurs, or when the re-searcher does not achieve to observe the predicted behavior, thesimulation must be restarted with other settings or constraints. Thisresults in the waste of an important number of compute cycles, assome simulations last for a long time: several days to weeks maybe required to reproduce a short timespan, a few nanoseconds, of molecular reality. Moreover, several biomolecular processes, likefolding or large conformational changes of proteins, occur on evenlonger timescales that are inaccessible to current simulation tech-niques. It can thus be necessary to impose empirical constraints inorder to accelerate a simulation and to reproduce an experimentalresult in MD. These constraints have to be defined  a priori , render-ing it difficult to explore all possibilities in order to examine variousbiological hypothesis.A new approach allowing to address these problems has emergedrecently: Interactive Molecular Dynamics (IMD). IMD consists invisualising and interacting with a simulation in progress, and pro-vides user control over simulation settings in interactive time. Afew Molecular User Interfaces (MUI), such as  VMD  [Humphreyet al. 1996], offer such capabilities, providing methods to interfacewith a molecular simulation engine, for example  NAMD  [Phillipset al. 2005].Our goal is to extend the IMD approach to a broader range of simulation engines, as the use of a specific simulation sofware ormodel often depends on the studied biological system. We havethusdevelopedagenericandindependent library, called  MDDriver  ,which allows us to easily interface molecular simulation engineswith adapted visualisation tools through a network connection. Asa first step, we have rendered the calculation modules easily inter-changeable while keeping the existing  VMD  user interface as MUI.The  MDDriver   library enabled us to study the dynamic behaviorof a real biological system, Guanylate Kinase (GK), simulated bydifferent calculation engines and represented by models at varyingresolution. The results of this study convinced us that VR devicesoffer innovative functionalities such as interactive haptic controlduring the simulation, allowing the user to better understand dy-namic biomolecular behavior. As a natural extension to this work,we started to intensify the use of VR functionalities in a new inter-active and molecular dynamics application. The main motivationwas to provide a VR framework allowing to quickly prototype newfeatures and modules especially designed for the IMD approach ina VR context. Such applications would be difficult to realise withan end-user oriented package such as  VMD  because of its complexsofware architecture and its desktop oriented approach. We have  based our design on three developments: the  MDDriver   library, thevisualisation toolkit  VTK  , and the  vtkVRPN   library.  VTK   providesthe ability to quickly develop new visual renderings and user inter-actions;  vtkVRPN   is a library to properly integrate VR devices into VTK   through  VRPN  . 1  MDDriver   : a library to interface molecularsimulations and molecular user interfaces In the  VMD/NAMD  architecture, the IMD network protocol [Stoneet al. 2001] was developed in order to interface the MUI with theMD engine. However, the use of a specific simulation engine andMUI strongly depends on the studied biological system and onuser habits. Adding IMD capabilities to other simulation enginesand molecular models as well as to a variety of MUIs in addi-tion to  VMD  and  NAMD  enables a whole range of new possibili-ties in interactive molecular simulations. This approach allows usto address a larger user community working on molecular mod-eling and simulations, sometimes based on their own home-madesimulation engines. Following these motivations, we developeda generic and independent library, called  MDDriver  , inspired bythe  VMD/NAMD  approach. We have encapsulated the IMD proto-col in the  MDDriver   library, allowing a developer to easily adaptMUI code and MD code in order to extend them with IMD fea-tures. This interface provides functions for the exchange of specificdata structures over a network: atom positions and system energies,computed for each simulation step by the MD engine (server part),and user-applied forces on a selected atom set (see Figure 1; clientpart). This approach was tested, applied and improved by integrat-ing calls to the  MDDriver   library into the  GROMACS   simulationengine [Hess et al. 2008], thus rendering the simulation interactivevia a MUI. Only minimal changes in  GROMACS   were required (5additional lines of code). We have used  VMD  as MUI in order tostudy the molecular behavior of Guanylate Kinase (GK) using anall-atom model and a coarse-grained representation [Baaden andLavery 2007] with  GROMACS  . Then we have tested a home-madesimulation engine dedicated to molecular docking, which was alsoIMD-enabled. Figure 1:  MDDriver library for interfacing a Molecular Dynamicssimulation with a Molecular User Interface We insist on the fact that the  MDDriver   library was designed foreasy integration into any molecular simulation engine providingtime series of particle positions. Indeed there are many approachescapable of simulating the dynamic behavior of biomolecules, suchas lattice simulations, elastic networks, coarse grain models or evenquantum mechanical and semi-empirical methods. 2 Molecular visualisation with VTK The  MDDriver   library was a necessary first step before developinga VR application dedicated to IMD. It allowed us to simplify theinterface between MD simulation and MUI. The biological appli-cation described in more detail in the results section, has shown thatVR devices offer well-adapted interactive modalities, especially thehaptic modality, allowing a user to dynamically control a biomolec-ular process and thus gain understanding of its intrinsic properties.As an extension to this work, we have developed our own interac-tive molecular user interface approach, based on the object-orientedvisualisation toolkit  VTK  , including the integration of VR devices.The open source project  VTK   is supported by a large user commu-nity. We chose this toolkit because of its high-level functionalitiesfor scientific vizualisation, and of its large user base. The mainmotivation for this approach is to provide a VR framework allow-ing to rapidly prototype new features and modules especially de-signed for controlling molecular simulation engines in interactivetime.  VTK   provides bindings to script languages such as Pythonand TCL, commonly used by biostructural researchers. Finally, VTK   makes it easy to implement standard molecular visuals suchas atom, ”ball and stick”, ribbon, or cartoon representations.In Figure 2 we show how to use a VTK pipeline in order to visu-alise and interact with a dynamic ”ball and stick” representation.Each sphere represents an atom (ball) and each cylinder a chemicalbond between two atoms (stick; see Figure 5). The  vtkPDBReader  classreadstheinitialmolecularconformation(atomtypes, atompo-sitions, and chemical bonds between atoms) that was obtained byexperimental studies and stored in the PDB molecular file format.It is then transformed into a  vtkPolyData  instance, designed in  VTK  for describing polyline or polymesh data (points and topology). Onthe one hand, a  vtkGlyph3D  instance filters points (atom positions)of this  vtkPolyData  instance considering each one as a sphere, us-ing  vtkSphereSource . It deals with the color and radius mappingaccording to atom type. On the other hand, a  vtkTubeFilter   instancefilters lines (chemical bonds) considering each one as a cylinder.The  vtkPolyDataMapper   translates these results into  OpenGL  com-mands. Other common molecular representations, such as ribbonsor secondary structures, and less used ones, like volume renderingand isosurface representations for illustrating the electrostatic po-tential around the molecule, can also be implemented using native VTK   classes. Figure 2:  Molecular dynamics ”ball and stick” representation us-ing the VTK graphical pipeline and IMD control events The dynamic behavior of this representation is computed by amolecular simulation engine, interfaced via the  VTK   client usingthe  MDDriver   library. Dynamic events trigger the position updateof all atoms. Timer events trigger the geometry update and thegraphical rendering.  3  vtkVRPN   : A library for integrating VR de-vices in  VTK  The  VTK   toolkit offers stereo rendering, but does not provide othernative VR functionalities, such as tracker or haptic device manage-ment or graphic clustering. We can cite the approach proposedby [Kok and van Liere 2007], extending  VTK   to a VR contextin a platform-independent and sofware-independent way. Anotherapproach proposed by [Shamonin ] is dedicated to add CAVE-like display functionality to  VTK  . Our approach is based on  VRPN  [Taylor II et al. 2001], a library managing many different VR de-vices.  VRPN   is used to handle actual user hardware and soft-ware settings related to VR device management.  VRPN   is prob-ably the most widely spread VR framework among biostructuralresearchers. It is also used by the  VMD/NAMD  tools. The  vtkVRPN  library was developed to interface  VRPN   and  VTK   in order to en-able the efficient use of VR devices within  VTK  . This allowed usto easily integrate all the devices supported by  VRPN   with littlechange to any existing  VTK   application. The  vtkVRPN   library iscomposed of two layers. The first one, the  VTK Encapsulationlayer  , encapsulates  VRPN   into  VTK   converting  VRPN   messagesinto VTK events. The second layer, the  Manipulator layer  , is com-posed of classes reacting to  VTK   events. The Manipulator layerallows to interact with graphical objects such as a 3D cursor ora hand avatar ( vtkWandManipulator  ) as well as with the camera( vtkCameraManipulator  ) in the  VTK   scene. The ( vtkWandManip-ulator  ) manages also haptic feedback, using high level VTK func-tionalities for simple feedback, and native  VRPN   capabilities forcomputationally demanding feedback. Figure 3:  Software architecture and VRPN encapsulation into VTK through the vtkVRPN library 4 Results In the following two studies, we have used the VR framework thatwas described in the previous sections. The use of a 3DOF trackeror an analog device allows us to control the camera. The use of ahaptic device controls the direction and amplitude of the forces ap-plied to selected atoms. Moreover the haptic feedback offers thepossibility to adjust the amplitude of the force according to theuser’s preferences. The interaction paradigm to interactively im-pose forces on particles contains two stages. The first stage is theselection of a single particle or a set of particles using a 3D cur-sor attached to a tracker device and its buttons. In a second stage,we use the force model described in [Stone et al. 2001] in orderto compute the force applied to the selected atoms. The resultingforces are rendered by haptic feedback if a haptic device is used,or by visual feedback such as the blue arrows shown in the top leftpart of Figure 4, and are simultaneously sent to the simulation en-gine using the  MDDriver   library.We first validated this approach on a simple test system, a polypep-tideofseveralhundredatoms, withandwithoutsurroundingsolventmolecules (about a thousand additional atoms; see Figure 4 below). Figure 4:  Dynamic haptic control of a simple polypeptide with(right) or without (left) solvent - ”ball and stick” representation In the second study, we have worked on a real biomolecular system,the Guanylate Kinase (GK) enzyme. Structures for this moleculeare provided by experimental methods such as Nuclear MagneticResonance or X Ray cristallography. The molecule has a U shapewith either a closed or an open conformation (see 5). The closuremechanism of GK consists in increasing the proximity of two sub-strate binding sites, for GMP and ATP, both essential for the en-zymatic reaction. The goal of our study is to understand whichparts of this system are involved in the closure mechanism. Thismechanism has been investigated using our  MDDriver   framework (VMD/MDDriver/GROMACS) at two levels of detail. The firstlevel corresponds to an all-atom model (18098 atoms), the secondto a lower resolution coarse-grain model (1900 beads). Prospectivetests using coarse-grain simulations allowed for the efficient explo-ration of a broad range of possibilities to close the enzyme, tryingto reach a closed conformation similar to the available experimen-tal structures. Figure 5 shows a secondary structure representationof the protein, considering specific architectural units such as theloops (white tubes), the helices (purple ribbons) and the beta sheets(yellow arrows). The crucial role of one loop (highlighted in red inFigure5)intheinitiationofGK’sclosurecouldthusbeidentified. Itwas then confirmed in a second phase using more detailed all-atomsimulations. Understanding the features of this early intermediatestate occurring as an impulse for the closure mechanism allows usto propose a novel mechanistic hypothesis. The loop move couldbe initiated by GMP docking, which may drive this loop via longrange electrostatic interactions. When the loop draws closer to the  other side of the enzyme, conformational changes could be trig-gered, subsequently inducing a global closure of the enzyme. Theinteractive exploration of the simulation using the haptic modalitylead us to this theoretical hypothesis. It also suggests that electro-static interactions could be the main driving force for closure. Figure 5:  Haptic control (red arrows in the red loop) of GuanylateKinase closure. Secondary structure cartoon representation of theopen state (left) and the closed state (right) Discussion The work presented in this paper is the first milestone in a largerproject currently in progress. Here we present a VR framework architecture, including a  VTK  -based molecular UI,  vtkVRPN   for in-tegrating VR devices into  VTK   and a library developed and testedin order to easily interact with various simulation engines such as  NAMD  or  GROMACS  . The framework is able to interact with a va-riety of molecular UIs, such as  VMD  or our own interactive molec-ular UI based on  VTK   and  vtkVRPN  . We have used this approachto investigate the dynamic behavior of the Guanylate Kinase en-zyme. The use of a simple haptic paradigm allowed us to interactwith different molecular simulation engines varying the level of de-tail of the molecular model and accessing different timescales. Thisinvestigation lead us to a new hypothesis on GK’s closure mecha-nism. Moreover, simulating and interacting with a molecular sys-tem based on different modelling scales allows to collect cooper-ative data and provides information about accuracy, pertinence orimplementation of simulation methods in terms of reliability of en-ergies and physical or chemical properties. Finally, the IMD ap-proach appears as a powerful tool to improve new simulation meth-ods under development - such as the coarse grain representationused in this work - or assess the impact of simulation parameters.It has to be noted that the time scale of a simulation, particularlywith respect to VR investigations in interactive time, imposes cer-tain limits. In order to produce an effective event on a dynamicmolecular structure, it may be necessary to impose an importantforcebeyondtherangeofpertinentbiophysicalenergies. Thisprob-lem needs to be addressed in future work. Another bottleneck con-cerns performance. While  VTK   has a well-adapted visualisationmodel for static molecular visualisation and interaction with thescene via high-level functionalities, a moving scene imposes addi-tional constraints. We are now working on increased rendering per-formance, compensating for the dynamic aspect of IMD which iscurrently limiting the interactivity to several thousand atoms. Thislimitation can probably be overcome by separating the simulationsystem into sub-parts. One sub-part would concern about 80% of all atoms, including solvent, surrounding the studied biomolecule.These particles are important for a pertinent biomolecular simu-lation with respect to rendering, but not for interaction. We areworking on the optimization of   VTK  , using a classical graphical VTK   representation for atoms of the sub-part concerned by the in-teraction, and more powerful techniques like shaders for the solventatoms, in order to be able to handle and interact with bigger molec-ular systems. Our objective is to deal with at least several hundredthousand atoms. Another interesting aspect of this work is the wayuser interaction is mediated to the simulation engine. We currentlyuse forces, as they are a natural way to impose external constraintson an MD simulation. However, sending forces to the simulationengine might not be the only paradigm for interacting with the sim-ulation. Othermodalitiessuchasimposingchangingatomiccoordi-nates can be considered. This approach should be straight-forwardfor applications such as interactive molecular rigid body docking,where the goal is to study the dynamic assembly and interaction of two rigid proteins, by applying global modifications to the atomicpositions of the selected molecule. We are currently investigatingthis approach in our in-house coarse grained docking software. Acknowledgements This work is supported by two ANR grants for software devel-opment (Project ANR-07-CIS7-003) and biological applications(Project ANR-06-PCVI-0025). We like to especially thank Jean-Phillipe Nomin´e, Gregory Journ´e, and Claire Guilbaud for access to the  vtkVRPN   library, which was developed at the ”Commissariat`a l’Energie Atomique”. References B AADEN , M.,  AND  L AVERY , R. 2007. There’s plenty of room inthe middle: Multiscale Modelling of Biological Systems.  Recent  Advances in Structural Bioinformatics , 173–196.H ESS , B., K UTZNER , C., V ANDERSPOEL , D.,  AND  L INDAHL ,E. 2008. GROMACS 4: Algorithms for Highly Efficient, Load-Balanced, and Scalable Molecular Simulation.  Journal of Chem-ical Theory and Computation 4 , 3, 435–447.H UMPHREY , W., D ALKE , A.,  AND  S CHULTEN , K. 1996. VMD -Visual Molecular Dynamics.  Journal of Molecular Graphics 14 ,33–38.K OK , A. J.,  AND VAN  L IERE , R. 2007. A multimodal virtualreality interface for 3D interaction with VTK.  Proceedings of Knowledge and Information Systems 13 , 2, 197–219.P HILLIPS , J. C., B RAUN , R., W ANG , W., G UMBART , J.,T AJKHORSHID , E., V ILLA , E., C HIPOT , C., S KEEL , R. D.,K ALE , L.,  AND  S CHULTEN , K. 2005. Scalable Molecular Dy-namics with NAMD.  Journal of Computational Chemistry 26  ,1781–1802.S HAMONIN , D. VtkCave : http://staff.science.uva.nl/ dsha-moni/myprojects/VtkCave.html.S TONE , J. E., G ULLINGSRUD , J.,  AND  S CHULTEN , K. 2001. Asystem for interactive molecular dynamics simulation.  Proceed-ings of Interactive 3D Graphics , 191–194.T AYLOR  II, R. M., H UDSON , T. C., S EEGER , A., W EBER , H.,J ULIANO , J.,  AND  H ELSER , A. T. 2001. VRPN: A Device-Independent, Network-Transparent VR Peripheral System.  Pro-ceedings of Virtual Reality Software and Technology , 55–61.
Search
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks