|Home | About | Journals | Submit | Contact Us | Français|
February 25–26, 2010, in San Francisco, the Resource for Biocomputing, Visualization and Informatics (RBVI) and the National Center for Macromolecular Imaging (NCMI) hosted a molecular animation workshop for 21 structural biologists, molecular animators, and creators of molecular visualization software. Molecular animation aims to visualize scientific understanding of biomolecular processes and structures. The primary goal of the workshop was to identify the necessary tools for: producing high quality molecular animations, understanding complex molecular and cellular structures, creating publication supplementary materials and conference presentations, and teaching science to students and the public. Another use of molecular animation emerged in the workshop: helping to focus scientific inquiry about the motions of molecules and enhancing informal communication within and between laboratories.
Several excellent accessible software packages have already lowered the barriers to visualization of molecular and cellular structures by structural biologists. Now the challenge is to provide a similar level of accessibility for animating the larger scale motions of those structures. Several excellent but not-so-accessible software packages exist for animation, as witnessed in the current explosion of sophisticated computer-driven Hollywood animations. Workshop participants are engaged in synthesizing these developments in molecular graphics and animation, through importing molecular graphics capabilities into animation packages developed by the cinema and entertainment industry, through providing animation functions in molecular graphics packages, through developing interfaces that can work with both types of software, and through interfacing animation and molecular graphics software with molecular dynamics, normal mode analysis, energy minimization and other physics-based simulation tools. A number of participants reported how animation moved their biomolecular research forward and facilitated communication among collaborators and the audience. Our report groups presentations made during the workshop into several categories: Needs and Requirements for Molecular Animations, Examples of Current Molecular Animation Technology, and Tools for use with Professional Animation Packages.
Nick Woolridge, director of Biomedical Communications at the University of Toronto (http://brodel.med.utoronto.ca/bmc/), spoke as an advocate for molecular animation from the point of view of visual cognition. He reminded the workshop that more than 40% of the human cerebral cortex is devoted to vision, more than all the other senses combined (Ware 2004). Molecular animators need to be knowledgeable in visual logic, and can profit from the visual best practices of film design and the “grammar of film.” For example, in a scene showing two actors in conversation, film makers use the axis between the two actors to define a limit for camera placements. They know that moving the camera from one side of this axis to the other will weaken the viewer’s grasp of the spatial relationships in the scene. Molecular animation may need additional, specialized rules. He drew the workshop’s attention to the importance of story-boarding. Specifically, a scientist may wish to tell a story using molecular animation, but the story may initially be conceived in a verbal mode. To plan an animation that tells a story well, the first step must be to create sketches of the frames at which changes are initiated and completed. This forces visualization.
Junjie Zhang, presently a postdoc with Michael Levitt at Stanford, spoke about molecular animations that he and Matt Dougherty created while Junjie was a graduate student with Wah Chiu at the NCMI, studying the mechanism of closing the folding chamber in a group II chaperonin (Zhang 2010). Animation moved this research forward at three points. First, it was necessary to animate the change between the open and closed states (which he determined using cryo-EM) in order to demonstrate that the mechanism is mostly a simple rocking of the subunits, a mechanism very different from previously studied chaperonins. Second, animations that visualized the structural changes in subunit interactions at the N and C termini effectively communicated to collaborators who were immediately able to connect what they saw to differences they observed in the protection of the termini to enzymatic proteolysis. A third animation showed the changes in distance between charged residues due to the conformational change, which led to mutational studies.
Zhang discussed various strategies for visualizing molecular motions. For simple changes, such as flipping of aromatic side chains, it is sufficient merely to switch between static images of the two states. For large conformational changes, such as the rocking motion of the chaperonin subunits, morphing between start and end states is effective. A more physical approach to large conformational changes is to apply normal mode analysis (in torsion angle space). This approach limits the degrees of freedom, making it possible to explore the flexibility of large molecular complexes. Finally, by breaking molecules into parts and allowing only those parts to move, a standard mature force field such as Amber can be used to explore the conformational freedom of large molecular machines.
Zheng Yang, a postdoc with Tom Ferrin at the University of California at San Francisco, discussed strategies for examining protein dynamics. For large conformational changes, such as the allosteric transition of chaperonin GroEL-GroES, morphing between end states may traverse unrealistic pathways. For generating more physically meaningful trajectories between distant pairs of functional states, coarse-grained methods, such as normal mode analyses (NMAs), may yield computationally efficient solutions (Yang 2009). NMA trajectories filter out high frequency “jiggling” and “wiggling” vibrations and thus highlight biologically interesting results. Gaël McGill, in collaboration with Mark Bathe at MIT, is currently integrating NMA visualization tools into Molecular Maya (discussed below).
Jie Fu, a postdoc from Joachim Frank’s lab at Columbia University, presented a research talk about transfer-messenger RNA (tmRNA) that poses a specific challenge for molecular animation. Using cryo-EM to generate native and mutant “single particle reconstruction” structures of tmRNA interacting with a ribosome, and informed by other biochemical data, she has generated a hypothesis for a complex topological feat performed by tmRNA in the process of releasing a stalled ribosome from mRNA. Partial unfolding of a pseudo-knot in the transfer-messenger RNA within the confines of the surrounding ribosome may be required to achieve the conformational changes observed with cryo-EM. Molecular animation could aid in visualizing this complex process. This proposition has stringent requirements: the unfolding needs to be consistent with what is known about RNA unfolding without violating steric limitations. The animation should not show more atomic detail than is implied by the hypothesis. This challenge met with a variety of suggestions for visualizing the process, including steered molecular dynamics simulations and augmented reality interactions with models (see Old and New Computer Visualization Technologies, below).
Peter Rose, Scientific Lead at the Protein Data Bank (PDB) in San Diego, spoke about the growing importance of molecular animation to the PDB’s outreach and education missions (http://www.pdb.org/). As data sets deposited at the PDB increase in complexity, the need for animations to showcase molecular machinery increases. Presently, the PDB website offers visitors online molecular viewers with several mini-animation tools. The Ligand Explorer interface can fly in ligands to their binding sites, and turn on hydrogen bond representations. The Simple Viewer can rock, roll, and rotate structures. The site also offers the Molecules in Motion Kiosk viewer, a full screen animation program, which loops through preprogrammed routines displaying user-selected structures from different perspectives and highlighting ligands in their binding sites. The PDB wish-list for future web accessible animations includes tools that explore interactions in ligand binding; that compare apo- vs holo-, and active vs inactive structures by morphing; that visualize enzymes binding ligands, catalyzing reactions, and releasing products; that show association and dissociation of complexes; that explain molecular machines; and that give multiple scale views of large structures and peel away layers to see inside them. An ultimate goal is to display molecular structures in their cellular contexts.
Matt Dougherty of the NCMI gave a systems analysis of the production process for scientific animation, emphasizing the importance of integration. Good communication between scientists and animators, combined with the animators’ ability to independently perform data verification are essential integration methods. The animations produced at the NCMI (http://ncmi.bcm.edu/ncmi/movies/) are used for scientific exploration, for conference presentations, and for supplements to journal articles. He also spoke for the need to establish data standards, for the importance to co-develop best practices, and for the formal peer review of animations, whereby the best work is recognized and awarded.
Jeremy Swan, biological visualization coordinator for the Division of Intramural Research in the Eunice Kennedy Shriver National Institute of Child Health and Human Development (http://dir.nichd.nih.gov/), updated the workshop on relevant events in and out of the NIH. He advocated for online scientific animation user groups such as the NIH Biological Visualization Interest Group (http://bvig.nichd.nih.gov) to help build community, and also for broadening the accessibility of online animations by creating audio descriptions and posting storyboards and transcripts of narration. He reported that videos uploaded to YouTube can then be machine translated and transcribed using YouTube technology. Users can download the resulting transcript, edit it, and upload the corrected version. He also reported that the American Society for Cell Biology (ASCB) is building a new open-source library of biological images and animations. Through the funding of an ARRA grant by NIGMS, the ASCB Image Library hopes to pioneer the field of bioimage informatics and make the library as accessible as the PDB and Genbank.
Art Olson, director of the Molecular Graphics Laboratory at Scripps (http://mgl.scripps.edu/), demonstrated something old and something new. He showed a complete animation from 1980, narrating from the perspective of 2010. Olson’s historic animation was shot, frame by frame, with a Bolex 16 mm movie camera trained on the video screen of an Evans and Sutherland vector graphics display, showing line drawing representations of the Tomato Bushy Stunt Virus (TBSV) crystal structure. The choreography of the movie demonstrated the symmetries and interactions of the viral coat subunits. In the vector representation of the time, there were no issues of lighting, hidden surfaces or rendering, but the issues of conveying a science story (e.g. showing transitions, integrating schematics with data, discriminating between concerted versus propagated transitions) were the same as they are now.
Olson demonstrated new uses of augmented reality and proprioception in understanding molecular events. In this process, the user manipulates physical objects upon which a computer generates 3D representations of molecules (O’Donoghue 2010). The result is that the user interacts directly with the molecular representation, without the intervention of mouse and keyboard. This method could be applied to threading Jie Fu’s transfer-messenger RNA (discussed above) as it unfolds inside the ribosome, for example.
Christina Johnson, artist/modeler and computer graphics manager of North Dakota State University’s Virtual Cell animated online textbook project, coaches and collaborates with undergraduate students to produce animations for education and outreach (http://vcell.ndsu.edu/animations/). Student animators train themselves in Maya (10–15 hours per week) so they are ready to start production after about 6 weeks. Virtual Cell is directed by North Dakota State professors who focus on producing animations for concepts their students find difficult to grasp. Virtual Cell research shows that students learn best from narrated animations after they have read the corresponding text. Subtitles compete with animated objects for visual attention, while narration complements the animation. Virtual Cell representations of biomolecules are simplified shapes (based on PDB information) and the processes are simplified to the undergraduate level. Simplification allows the animators the freedom to portray events that are not well understood, and to size objects out of scale for clarity. All animations are done in “one shot” (without cuts) because every cut potentially confuses some viewers. Virtual Cell employs 3D animation because it can be perceived as a physical event, involving the viewers attention, while 2D animation remains more abstract and less engaging.
Anastasia Aksyuk, a postdoc in the Rossman lab at Purdue, described the use of animation to demonstrate a rigid body conformational change in a crowded molecular environment. Anastasia proposed a path for the large scale rearrangement of subunits in the infective contraction of the tail sheath of bacteriophage T4, based on fitting x-ray crystallographic structures of sheath protein deletion fragments into the low resolution cryoEM density maps of the assembled tail. Animation with Chimera showed that it is possible for the outer tail sheath subunits to complete the complex rearrangement implied in the structural data without remarkable steric clash, by simultaneously tilting the outer domains of the tail sheath proteins as rigid bodies and moving the subunits radially, rotationally, and longitudinally with respect to the tail sheath axis. While there does not appear to be much steric collision to hinder the process, the next step for this animation study is to incorporate steric constraints and electrostatic properties in the animation process.
Hang Yu, a graduate student in Klaus Shulten’s lab at University of Illinois at Urbana-Champaign, presented a tour of VMD (http://www.ks.uiuc.edu/Research/vmd/), a molecular visualization program for large biomolecular systems. VMD is designed to interface with molecular dynamics programs, to the point that it can be used to concurrently display and interact with a running NAnoscale Molecular Dynamics (NAMD) simulation. It displays a very wide range of data types, including amino acid and nucleotide sequence data, molecular coordinate trajectories, and volumetric data, as well as electrostatic and quantum mechanical data. VMD can be accelerated with Graphic Processing Units (GPUs), making it efficient for visualization and analysis of very large scale simulations. The built-in VMD movie making tool enables automatic creation of basic movies of trajectory animation, structure rock and roll, and rotations. More complex VMD animations can be created by writing scripts in the Tcl or Python programming languages. In addition to supporting interactive rendering and movie making with GPUs, VMD is also closely integrated with the Tachyon parallel ray tracing package, and can export molecular scenes to other photorealistic rendering tools including POV-Ray, Renderman, and others.
Tom Goddard, Programmer/Analyst at the RBVI, focuses on enabling the “everyday molecular structure researcher” to make, in Chimera (http://www.cgl.ucsf.edu/chimera/), simple molecular videos that spin, or morph, or zoom in on parts of structures. He demonstrated how videos that rock or rotate structures allow the viewer to perceive the 3D geometry of interacting protein residues far better than 2D static images, or even 2D stereo images. Understanding how structures differ is facilitated by simple linear morphing. Even with an unordered series of NMR structures, morphing allows the viewer to distinguish twist, bend and other large scale motions.
Goddard presented an informal but exhaustive survey of online movies supplemental to structural biology articles published by a high profile journal in the past year. About 5–10% of the articles on structural models included supplemental videos. All but 2 of the 17 articles showed simple video clips of rotating, morphing, or slicing structures, but none exploited the power of animated visualization to give much insight.
An obstacle to creating molecular animations in applications such as Chimera is the need to learn the necessary scripting commands. Morphing, however, can be done with a one-button command, and it would be good to incorporate that kind of functionality in Chimera for flying objects in and out of the frame, and for rotating, and rocking and rolling structures.
Another obstacle is the frustrating lack of standardization and reliability in the online technology for viewing computer animations. Ideally, molecular animation should be embedded in online articles, just as figures are. Journals typically don’t provide embedded animations because it is too difficult. Only YouTube currently has a reliable solution that doesn’t entail downloading and installing specialized software. However, because the current content of structural biological videos is lagging, there is little motivation for journals to spend the money to match YouTube’s capabilities. The part of the solution that is in our hands is to make 3D molecular animation sufficiently accessible to scientists that they can invest their efforts in making videos that communicate well, rather than spending inordinate amounts of time learning to write scripts.
Maya is an advanced and proven Hollywood film industry toolset for producing sophisticated 3D animations (e.g. Toy Story and Avatar). It simulates the behavior of materials such as cloth draping, fluid flow and particle collisions with a Python language scripting interface. Unfortunately, in addition to Maya’s steep learning curve, the software was not designed to incorporate molecular information. Although Protein Data Bank (PDB) data can be imported into Maya as a set of spatial positions for particles, the software ‘out-of-the-box’ knows nothing about bonds, atom types, secondary structures, hydrophobicity or electrical charges (McGill, 2008). To overcome these obstacles, Gaël McGill, Director of Molecular Visualization at Harvard Medical School and CEO of Digizyme (http://www.digizyme.com/), is developing Molecular Maya (mMaya), a free open source Python module that plugs into Maya. This module incorporates molecular properties so that Maya’s sophisticated engines for moving complex structures and rendering photorealistic images can be applied. In addition to importing and animating PDB datasets, the ultimate goal of mMaya is to allow scientists and animators to merge structural, dynamic and other biological data to create increasingly accurate molecular and cellular visualizations. mMaya is provided with detailed tutorials on the Molecular Movies web portal (http://www.molecularmovies.com/toolkit/index.html).
Michel Sanner of the Scripps Molecular Graphics Laboratory (http://mgl.scripps.edu/) presented various approaches for specifying and rendering molecular animation. In his molecular visualization program, Sanner developed several different types of user interfaces, which are rooted in modular programming principles and function with basic OpenGL laptop molecular graphics. He found that animation scripting, visual programming, and keyframe/timeline manipulation each quickly become too complicated for most users. An experimental PowerPoint-like interface proved to be too limited. A more satisfactory approach was for users to input a series of scenes (snapshots of program states) created in the PMV molecular graphics package, along with specifications of the time intervals between the scenes, into his Scenario software. Scenario then calculates the differences between sequential pairs of scenes and determines whether transforming the visualization from scene to scene requires it to recolor or change object representations, or to rotate, translate or morph objects. Scenario then calls an animation engine to render the intervening frames. In parallel with efforts to augment their Python Molecular Viewer (PMV) with animation capability, Sanner and Ludovic Autin developed ePMV, a version of PMV that can be embedded in professional 3D rendering and animation software, effectively extending them with support for biomolecular structural data. The ePMV solution thus leverages some of the best tools for 3D rendering and animation. This approach is feasible for any 3D graphics package that embeds a Python interpreter. It has been demonstrated for Maya, Cinema4D and Blender.
Monica Zoppè, director of the Scientific Visualization Unit of the Institute of Clinical Physiology at the CNR (National Research Center) of Italy (http://www.scivis.ifc.cnr.it/), spoke about a molecular animation process that proposes to use computer graphics tools to obtain navigation maps for protein motions, and also discussed issues of representing lipophilic and electrical forces. She advocated for the use of the Blender animation package because it is free and open-source. She described the process for obtaining intermediate frames for the conformational transition between different models of apo-calmodulin (Kuboniwa 1995). The RMSD among NMR structures from the PDB are calculated, and the two most disparate structures are chosen as endpoints for the transition. The conformations are imported into Blender as hard spheres with defined collision radii and with bonds set as rigid body joints. The Blender game engine then very rapidly calculates a transition between the two conformations and outputs a set of proposed intermediates. This set is compared with the NMR data, and the conformations within 2 Å RMSD of a Blender transition conformation are selected and ordered according to the sequence of the Blender transition. The process is then repeated to determine the transition between the next pair of most disparate conformations, continuing until all the experimentally determined conformations are ordered. The energies of the conformations are minimized using the GROMOS force field implemented in Swiss-PdbViewer (Guex 1997). When this process is applied to each possible pair of conformational endpoints, the result is a navigation map.
Zoppè proposes a new scheme for representing surface features, which, she claims, are naturally perceived as physical properties of matter and convey a more immediate understanding of molecular properties. The proteins are represented as surfaces, textured according to the local hydrophilicity. Hydrophilic regions are dark and rough and hydrophobic regions are smooth and shiny. Electric fields of proteins are calculated using the Adaptive Poisson-Boltzmann Solver module and represented by Blender particle emitters placed along field lines. The emitted particles travel along the field lines from positive to negative. The goal is to encode the invisible molecular world in visually intuitive form. Portraying temperature, pH, the crowdedness, and the time and spatial scales of the cellular world are tremendous challenges.
The workshop ended with a brain-storming session on strategies for facilitating, improving, and supporting molecular animation. The discussion was swift and wide-ranging. Ideas that came up repeatedly included a need for: a moderated online molecular animation community forum and repository for animations, scripts, user ratings and comments; peer and jury review of animations; hands on workshops in Maya, Blender, Chimera, etc.; and workshops for collaborations between biologists and animators.
A topic that repeatedly arose during discussions was what to do about missing data, especially when trying to represent dynamic processes for which experimental data is only available for starting and ending conformations. For example, the details of the complex conformational changes that transfer-messenger RNA undergoes in the studies Jie Fu described above are not completely known, but constructing animations depicting possible topological pathways can provide useful insight and lead to formation of new hypothesis and ideas for experiments. How can the portion of the animation that accurately represents high-quality experimental data be differentiated from the portion that depends on computationally modeled data, or portions dependent on hypothetical data only? Similarly, if some of the structural details involved in a molecular process are unknown, when is it acceptable for an author to use artistic license to fill in the missing details? The resulting animation may be valuable for use in an educational environment, even if not all of the details are scientifically accurate. These and other challenges provide the motivation for much future work in the field.
We thank the workshop attendees for their participation and John “Scooter” Morris for his work in recording, editing, and posting videos from each talk. Recorded PowerPoint presentations are accessible at http://plato.cgl.ucsf.edu/trac/Workshops/wiki/AnimationWorkshop. Many of the participants also posted PDFs of their slides and links to sample animations of their work; these are accessible via this same URL. The meeting was organized through the National Center for Macromolecular Imaging and the Resource for Biocomputing, Visualization and Informatics, with support from the National Center for Research Resources (P41RR002250, P41RR001081) and the Nanomedicine Development Center (PN2EY016525).