Waldon et al. BMC Bioinformatics 2014, 15:334 http://www.biomedcentral.com/1471-2105/15/334 RESEARCH ARTICLE Open Access SketchBio: a scientist’s 3D interface for molecular modeling and animation Shawn M Waldon*, Peter M Thompson, Patrick J Hahn and Russell M Taylor II Abstract Background: Because of the difficulties involved in learning and using 3D modeling and rendering software, many scientists hire programmers or animators to create models and animations. This both slows the discovery process and provides opportunities for miscommunication. Working with multiple collaborators, a tool was developed (based on a set of design goals) to enable them to directly construct models and animations. Results: SketchBio is presented, a tool that incorporates state-of-the-art bimanual interaction and drop shadows to enable rapid construction of molecular structures and animations. It includes three novel features: crystal-by-example, pose-mode physics, and spring-based layout that accelerate operations common in the formation of molecular models. Design decisions and their consequences are presented, including cases where iterative design was required to produce effective approaches. Conclusions: The design decisions, novel features, and inclusion of state-of-the-art techniques enabled SketchBio to meet all of its design goals. These features and decisions can be incorporated into existing and new tools to improve their effectiveness. Keywords: Molecular modelling, Animation, Collision detection Background both slows the discovery process and provides opportuni- SketchBio is a new tool to help scientists think about 3D ties for miscommunication. This paper describes an effort molecular structures and interactions and to communi- to provide scientists with a tool that is so rapid to learn cate them to others. and powerful to use that they can create these models and We found ourselves repeatedly using 2D hand-drawings animations themselves. of complex 3D structures and their interactions in discus- This tool should be general and widely useful. Many sions with our close collaborators in cell biology, pathol- researchers studying cell structure and physiology seek ogy, and chemistry, despite the fact that the 3D crystal to construct and evaluate dynamic models that incorpo- structures of the proteins making up these structures were rate random thermal motion as well as conformational known. Overall structure comprehension was advanced changes induced through intermolecular interactions. when a hired artist produced 3D scale models and com- Discovering, testing, and communicating hypotheses puter models of the structures [1]. Our group is not alone. about these interactions requires the development of Discussions among collaborators are often done using 2D complex animated 3D molecular structures. Modeling, whiteboard sketches. Presentations often consist of pasted simulation, and rendering these hypothetical scenarios images and 2D PowerPoint animations. involves using a number of tools and databases (PDB, Due to the difficulties involved in learning and using PyMol, Blender, NAMD, etc.) and then converting files 3D modeling and rendering software, many scientists hire to pass geometry and animations between tools. It also professional computer programmers and/or animators to involves manual placement and orientation of 3D objects, work with them to create models and animations rather which is currently done using 2D input devices and by-eye than use these programs themselves. This indirection detection and avoidance of collisions. As a result, it often takes a team months to produce an acceptable model or *Correspondence: [email protected] animation. University of North Carolina at Chapel Hill, 27599 Chapel Hill, NC, USA © 2014 Waldon et al.; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated. Waldon et al. BMC Bioinformatics 2014, 15:334 Page 2 of 17 http://www.biomedcentral.com/1471-2105/15/334 The aim was to produce a tool that reduces this to a The final driving problem involved cell division (mito- single person working for hours or days. sis). Many proteins beyond cohesin and condensin con- This paper describes that tool, SketchBio. tribute to mitosis. Scientists are able to fluorescently label both these proteins and chromosome locations and deter- mine relative distances and orientations between pairs Driving problems of proteins. With accurate localization and tracking for Fred Brooks posits that the best way to construct a tool 3D images, these techniques provide partial information that is generally usable is to focus on several very different on the 3D layout of proteins and chromosomes in wild- specific problems and build a tool that solves them [2]. type and mutant mitotic spindles. Building models to This approach was followed here. match this information requires the development of semi- The first driving problem for this project was to con- automatic layout of proteins. This will provide a partial set struct a protofibril model based on geometric constraints of constraints for scientists to construct protein-protein among a set of individual fibrinogen molecules. The pro- and protein-chromosome complexes that match experi- tein fibrinogen is the main component of blood clots, mental data. With these enhancements, SketchBio could where it is converted into fibrin and links together with be widely useful to other researchers for the generation other fibrin molecules to form strands. Two of these of hypothetical protein-complex structures from partial strands join together to form a protofibril, which form data. thick fibers that make up a large portion of the blood clot. Based on the crystallized structures of fibrin monomers Design goals from different species and on only two sets of known The application-specific needs from the above collabora- interactions [3], one collaborator sought to construct 3D tors can be summarized as a set of domain-independent protofibril structures matching those seen in her data, design goals for SketchBio: which suggested a structure in which two fibrin strands • twist around each other, and wanted to create a model Easy to learn and to use. Scientists must be able to that shows this interaction at the molecular level. Over rapidly construct models and animations on their several months, this collaborator and her students worked own using interfaces that enable them to concentrate with a computer scientist to use the powerful UCSF their mental efforts on the design challenge rather Chimera tool to construct such a model (“snapshots” than decyphering the interface. • and modeling of the early stages in fibrin polymeriza- Support molecular operations. It must be easy to tion, submitted). Building this model required repeated load molecules, extract the relevant substructures, iteration of hand-placement of two molecules (using mul- describe conformational changes, group molecules, tiple 2D mouse interactions), followed by using repli- and color according to standard data. cation tools to develop candidate models, which were • Appropriately constrain layout. Some molecular then evaluated against the data. The desired use of structures should not overlap, others (drug vs. SketchBio was to construct this protofibril rapidly and protein) overlap as part of their function, others semi-automatically by specifying which location on each (fibrin, actin) assemble into repeated structures. In fibrin should be in close contact with other molecules some cases, the distances between individual and by specifying that the molecules do not overlap. elements is known but their 3D layout is not. This same capability will enable generation of other Supporting all of these cases will enable a biologist to self-symmetric structures such as actin filaments and most rapidly explore the space of possible microtubules. conformations to produce consistent models. The second driving problem was to construct 3D mod- • Support rapidly iterated, in-context design. els and animations of the interaction between actin fila- Understanding the interactions between dozens of ments and vinculin. Actin filaments are one of the three molecules requires repeated adjustment of proposed main components of a cell’s cytoskeleton, and the pro- locations and motions. The reasonableness of tein vinculin binds to actin filaments, connecting them to interactions depends on nearby molecules, which other actin filaments or different proteins. change over time. Generating consistent models The third driving problem was to construct models requires trying and optimizing many potential of the mitotic spindle, a structure that separates chro- solutions before the final model is found. mosomes during cell division. As in the fibrin case, • Support high-quality rendering. Onceaproposed each step of model generation required support from an model has been completed, static and animated artist, animator, and/or programmer to convert a col- images that use the most-effective lighting and laborator’s concepts into geometry for rendering and surface rendering techniques are critical to conveying simulation. the model