Published in Chemistry in Britain, 1998, 34, 26.

Virtual Reality Modelling Language (VRML) in Chemistry

Omer Casher,a Christopher Leachb, Christopher S. Pageb and Henry S. Rzepa* b

aPresent address: Molecular Simulations Ltd, 240/250 The Quorum, Barnwell Road, Cambridge CB5 8RE England
bDepartment of Chemistry, Imperial College, London, SW7 2AY.

The number of chemistry-related World-Wide Web sites has grown from a few hundred since our introductory review in 19951 to around 3600 by early 1998.2 Much of the chemical content of these sites has been expressed using Hypertext-markup language (HTML), incorporating hyperlinks in the form of the now familiar URL (Uniform Resource Locator) to create associations with other documents and sites, and to reference visual content such as two-dimensional images, diagrams and schemes. Since 1994 however, a number of novel technologies have been introduced to the Web which go beyond the use of simple images. Here we will focus on one such method termed Virtual Reality Modelling Language or VRML, which has been applied in a number of chemically interesting ways.

The limitations of Images

The basic object collection used to construct a document in HTML is the ASCII character set which includes the letters you are reading now, together with some specially reserved control characters such as < or >. Together with some Greek symbols (which are actually not handled well in HTML), highly complex chemical meanings, semantics and data can be expressed. Chemistry however can be a particularly visual subject, and many of our models and data of molecular behaviour and structure are most easily comprehended and disseminated using visual means of expression. On the Web, most visual illustrations have hitherto been derived from bit-mapped digital formats, or images as they are known. As devices for expressing chemical content and meaning, such images offer little advantage over the use of print, and suffer from the same limitations such as the great difficulty in indexing of and searching for the meaning they carry. Another limitation (or advantage, depending on your point of view) is that such illustrations show only their author's interpretation and selected viewpoint of a particular chemical concept or expression of data and are subject to copyright control. The reader cannot select any other viewpoint or impose upon that viewpoint any other style, nor can they easily recover in an error-free manner any of the original data or information used to generate the illustration, or indeed copy the image without permission.

The Importance of Models

We believe that a superior approach lies in defining multi-dimensional models wherever practicable, rather than in creating static 2D illustrations. Such models could if needed have attributes of time-dependence (i.e. animation), sounds and behaviour controlled by specified algorithms. Virtual reality modelling language (VRML) was born at a workshop at the first World-Wide Web conference held in Geneva in May 1994 as an expression of this need, and even then chemical applications were envisaged.3 The first definitions of this new framework emerged in October 1994 and became standardised as version 1 during 1995. This was superceded by VRML 2 in 1996, and has now been renamed VRML 97.4 It is worth emphasizing that VRML was designed as a generic modelling language, in contrast to HTML, which is a markup language. Markup is a mechanism which allows the author to express semantics and can thus be used to provide fine grained structure and relationships in a document. Such internal structure in turn allows indexing of the content. VRML, as a modelling language, is currently less suited for semantic expression, and hence for operations such as indexing.

To illustrate the difference between a modelling and a markup language, consider how an atom might be described. HTML itself has no well defined mechanism for defining an atom, and one has to use a markup language such as CML5 (Chemical Markup Language, which is an implementation of XML, itself an evolutionary successor to HTML) to define an atom and properties such as atomic number, its connectivity to other atoms, the number of electrons associated with it and so forth. Using VRML, one would define the same atom as an spherical object, with model properties such as colour, radius, 3D co-ordinates, lighting and motion attributes, and if necessary associate this object with scriptable actions such as collision avoidance with other objects (computed if necessary using e.g. molecular mechanics force fields). VRML is therefore complementary to a markup language since it defines a quite different set of primitive and importantly three-dimensional objects which can be used to express complex chemical models in a visual manner. Copyright implications for models are also different from illustrations, in part because models are specified using data provided by their author, and only created in a specific viewable style by the actions of the reader and the software they are using.

The Characteristics of a VRML Model

The first thing one notices about a VRML model is that it must be navigated using a quite different set of metaphors from that used for a page of HTML-derived text. The dashboard controls (Model 1), which are probably more familiar to a rather younger generation of fast-action games console owners than to most chemists, contain some fascinating differences from printed documents and HTML-browser pages. Whereas a search-index or the "find" browser button for example can be used to easily locate occurances of a specific set of objects (characters) within a printed or HTML based document, no such option is currently possible in a VRML model. The best one can do in this regard is to "seek" an already visible object by changing the viewpoint of the model to approach it more closely. Another option found on a page of displayed text and images is the print button; an option conspicuously missing from VRML controls. This would imply that any electronic publication which might make extensive use of models to convey chemical information could not have any directly equivalent printed form. This very article is one such example; what you are viewing here of course are illustrations reduced from the information provided by models.

One page-derived metaphor does carry over; the author of a model can insert the equivalent of section headings (called viewpoints) to emphasise what they believe are important characteristics of their model, and in the process also provide the equivalent of keywords ("meta-data") to assist in any indexing of their model. A rather less desirable characteristic of VRML models is that even minor syntactic errors in their specification (the equivalent of spelling mistakes or errors in the markup of an HTML document) will probably render the entire model unviewable, a contrast with component-based HTML documents which fail more gracefully in failing to properly display only the component with the error. This intolerance of errors is probably one reason why the application of VRML is still relatively low.


Model 1. Navigation Dashboard for a VRML Model, deriving from the CosmoWorld viewer.

Viewing and Creating VRML Models

The widespread adoption of VRML has also been limited by the need to use relatively powerful computers. Only in the last year has the introduction of low cost, fast (>200MHz) computers equipped with powerful graphical capabilities born from the need to sell computer games, and adequate memory (> 32 Mbyte), coupled with the release of software such as CosmoPlayer6 made viewing VRML models a practical reality for the average computer owner. CosmoPlayer can view VRML 2.0/97 models and comes in the form of a plug-in for use with Web clients such as Netscape.

Creating VRML worlds is more complex than authoring an HTML document, and invariably requires the acquisition of suitable software. A good generic package is CosmoWorlds6, whilst for chemical applications, programs such as WebLab Viewer Pro7 or EyeChem8 are available. Several "filters" for converting standard chemical coordinate formats such as PDB also exist.9

The Application of VRML Models in Chemistry

1. Complex Molecular models

VRML has been applied to many areas of chemistry9. The first VRML models were published by our group in December 1994 (Model 2) resulting from our experiments10 in what we called "molecular collaboratories" using the newly introduced UK high speed national computer network. These models retained the familiar wireframe, ball and stick, spacefill or ribbon representations of molecules, but enabled the rich navigation features of VRML (Model 1), and the ability to define hyperlinks from individual regions of the molecule to other Internet resources, in much the same way as HTML allows such hyperlinks in text-based documents.
Model 2. An early VRML model showing the 3CRO protein/DNA complex.

Another great advantage of VRML is that it can be readily extended to displaying more complex chemicals such as ionic lattices, large biopolymers, carbohydrates, peptides, liquid crystals etc. These often require different symbolic representations to illustrate the key features. Some excellent examples of this type of application emerged from Brickmann's11 group in Darmstadt from 1995 onwards. Their work shows how more complex models (Model 3)11 can reveal the 3D structure of e.g. the p53 tumor suppressor protein complexed with a DNA helix (blue); the peptide backbone of the protein is rendered as a ribbon, with the computed electrostatic potentials at the interface being projected onto the respective Van der Waals surfaces of the molecules. The red groups are the "mutation hotspots".
Model 3. A VRML model illustrating the key features of the p53 tumor suppressor protein (taken from Y. Cho, S. Gorina, and N. Pavletich, Science, 1994, 265, 346).

Hewat in Grenoble12 showed early on how a VRML representation (Model 4) of the structure of the high-temperature superconductor Y2Ba4Cu7O15, composed of alternating Y1Ba2Cu3 and Y1Ba2Cu4 units could be constructed dynamically from a set of options and parameters provided by the user via a Web-based form.
Model 4. VRML representation of a high-temperature superconductor (taken from A. W. Hewat, P. Fischer, E. Kaldis, J. Karpinski, S. Rusiecki and E. Jilek, Physica C, 1990, 167, 579 ).

In principle, any computed molecular surface can be represented in VRML. Model 5 illustrates a model showing the HOMO orbital computed for a supermolecule corresponding to the peripheral light-harvesting complex (LH2) of purple photosynthetic bacteria.The structure includes two transmembrane polypeptides and associated pigments, a pair of closely interacting bacteriochlorophyll-a molecules (B850) and a carotenoid (Car) unit. Constructing such models enables detailed investigation of the method of energy transfer via electrons and protons in photosynthetic bacteria.
Model 5. VRML Model showing the Highest Occupied Molecular Orbital of a bacterial light-harvesting complex (G. D. Scholes, I. R. Gould and G. R. Fleming, to be published).

2. Knowledge Discovery and Data Mining

Bragg once said that "the important thing in science is not so much to obtain new facts, as to obtain new ways of thinking about them". Knowledge discovery in databases (KDD), or, more popularly, "data mining", has generated a great deal of interest in recent years as vast quantities of information have begun to accumulate in databases. The objective is to be able to trawl these databases in novel ways so as to discover unusual relations and correlations between and within the data. We have applied VRML13 to the problem of representing 3D information "mined" from e.g. the Cambridge Crystallographic Database14 to help identify the characteristics of weak intermolecular hydrogen bonding interactions around aromatic systems (Model 6). Visual representations can allow certain attributes in the data to be noticed precisely because they are unusual. A VRML-based model permits us to view different types of data collected together in one scene, and to make links between different levels in the data hierarchy to allow probing of any unusual facets that may emerge in a highly intuitive manner.
Model 6. A VRML cluster diagram showing intermolecular hydrogen bonding interactions around aromatic systems

The model shows the results of a search for short contacts of some typical hydrogen bond acceptor groups to a chlorobenzene ring, normalised against the Connolly surface, i.e. that portion of the van der Waals surface that is accessible to a probe of finite radius. Such an approach makes no presuppositions but rather reveals structural preferences, as is borne out by the fact that the vast majority of the contact vectors are oriented towards the ring hydrogens. The conversion to VRML allows for the vectors, colour coded according to the type of contact group, to be hyperlinked to a second model which in turn can be used to highlight the significant interaction, and give other information such as the unit cell and a literature reference. This reference in turn can be a link to the original electronic journal article which allows the research to progress from the discovery of a potentially interesting effect to reading the original article about it.

Another application of data mining is illustrated in the Panel, which shows how subtle structure-activity phenomena such as hydrogen bonding can be teased out of large amounts of numerical data provided by the technique of molecular modelling.15

3. Animation

Adding time-dependence to a model can often enhance the perception and understanding of a subtle chemical phenomenon. We experienced this directly when studying the potential energy surface of a sequence of pericyclic reactions starting at 1 or 2 and forming cyclo-octatetraene (4). After a conventional article had been prepared for printed publication,16 we decided to enhance it for an electronic version of the journal17 by including selected 3D models, and animating those models to show normal vibrational modes. Whilst proof-reading the "enhanced" article, it suddenly dawned on us that the animated mode for one geometry in particular had unexpected characteristics.

Ostensibly, the reaction corresponded to two synchronous electrocyclic ring openings proceeding in a particular manner known as conrotation. Each individual ring opening is, according to a theory first proposed by Woodward and Hoffmann, thermally allowed. From another perspective however, the reaction also corresponded to a cyclo elimination reaction thermally forbidden by the Woodward-Hoffmann rules. How could both interpretations be true? Focusing our eyes on the four atoms corresponding to the cyclo elimination, we realised by watching the animation that the Woodward-Hoffmann forbidden characteristics were in fact avoided by a pronounced lateral dislocation of the two reacting centres (Model 7). If you are reading this article in print, and are having difficulty visualising what we mean by that, we strongly urge you to view the animated VRML model to see how we obtained our original insight.18

Model 7. An animated model showing vibrational modes.

Simple animations of this type can also be accomplished using other tools for displaying molecular models such as Chime for molecules or spectra19, but this requires a set of molecular coordinates for every frame of the animation. VRML often requires only the starting and ending coordinates of the molecule. The properties of the various objects within a VRML scene are then interpolated at each time step to create the effect of an animation. Such a technique is particularly effective when animations of large molecules are required.

4. Virtual Laboratories

A quite different application of VRML illustrates how laboratory instrumentation or apparatus might be depicted (Model 8).20 Having a more "lifelike" rendition may make it easier for others to reconstruct a complex experimental set-up. It is also possible to make the model interactive to facilitate the appraisal of modifications to the scheme. For example, various VRML laboratory components or "objects" could be connected together with the help of a VRML authoring tool from an on-line "storeroom". The virtual approach would also provide a means for students to gain some insight to modern laboratory techniques that might otherwise be prohibitively expensive to actually carry out. Ultimately, a network of virtual instruments corresponding to the layout of real world laboratory environment could be constructed, and virtual experiments within the laboratories could correspond to live experiments.

In the scene shown here, each important object is assigned a "viewpoint" to help in the navigation, and these viewpoints could be used to index the model, and ultimately to search for model components on a global scale.

Model 8. A Virtual Molecular Laboratory, with hyperlinks to other information sources.

There is even the possibility that a real instrument could be depicted and even controlled remotely, in a manner analogous to the automated process controls in a chemical plant. An automated laboratory driven largely by robotics could be managed by an expert from practically anywhere in the world. The clear advantage here is that several such laboratories can be managed by the same person. If the VRML world reflects real world applications, many challenging issues arise. In the case of the virtual laboratory, a mechanism is required whereby the laboratory information is transferred to the VRML scene. If the laboratory information is stored in databases this would entail suitable integration between the VRML model and the database. This can be achieved using a technique known as VRML scripting. Here, little programs are embedded within the VRML model giving it behaviour characteristics. For example, one such program might give a simple VRML model of a clamp the ability to pivot, or in a molecular model, a program might be embedded to calculate a required property of the molecule.

Concluding Remarks

In describing their subject in print, chemists have hitherto had to rely on static two dimensional visual devices such as figures, diagrams and schemes. The introduction of tools such VRML and the World-Wide Web has added the dynamic "model" to this repertoire18. We anticipate that VRML or similar tools for constructing complex visual models will gradually become incorporated into mainstream electronic journals17, books and other electronic publications. As the chemical content carried by such models becomes richer, a major challenge for the future will be the development of effective means of indexing these models in a chemically meaningful manner, and creating a mechanism to search for specified components of such models on a global scale.

Acknowledgements

We thank SmithKline Beecham and Glaxo Welcome for studentships (to CSP and CL respectively) and the JISC E-lib programme for CLIC project funding (to OC).

References

  1. M. J. Winter, H. S. Rzepa and B. J. Whitaker, Chem. Brit., 1995, 685.
  2. M. J. Winter, ChemDex. See http://www.shef.ac.uk/chemistry/chemdex/welcome.html
  3. H. S. Rzepa, in " Selected Papers of the First World-Wide Web Conference", Comp. Net. ISDN Systems, 1994, 27, 317.
  4. Full details of the history and specifications can be seen at the VRML repository at http://www.sdsc.edu/vrml/. This site also contains a comprehensive list of available software and examples of VRML models.
  5. P. Murray-Rust, "Chemical Markup Language, A simple introduction to structured documents", in "XML, Principles, Tools and Techniques" (Ed. D. Connolly,), O'Reilly, 1997, pp 135-149. For details of all W3C (World-Wide Web Consortium) recommendations, proposed recommendations, working drafts and notes, see http://www.w3.org/TR/
  6. CosmoPlayer can be obtained from the CosmoSoftware Web site; http://www.cosmosoftware.com/download/player.html. This comes complete with extensive help pages, and VRML examples. The Macintosh version can also be obtained from here.
  7. M. Pear, J. Berstein, C. Z. Le, R. Mcdonald, Abs. Papers Am. Chem. Soc 1996, 212, 28-Cinf; J. Mol. Graphics Mod., 1997, 15, 62. See also http://www.msi.com/weblab/
  8. O. Casher and H. S. Rzepa, J. Mol. Graphics, 1995, 13, 268. Details of this group's work can be seen at http://chemcomm.clic.ac.uk/VRML/ together with useful tools; http://www.ch.ic.ac.uk/omer/vrmltools
  9. An excellent review summarises much of this work; W. D. Ihlenfeld, J. Mol. Model., 1997, 3, 386-402.
  10. O. Casher and H. S. Rzepa, Computer Graphics, 1995, 29, 52.
  11. G. Moeckel, M. Keil, M. Hollstein, B. Spiegelhalder, H. Bartsch, J. Brickmann J. Mol. Mod., 1997, 3, 382-385; G. Moeckel, M. Keil, B. Spiegelhalder, J. Brickmann, ibid, 1996, 2, 370-372. Details of this group's work can be seen at http://www.pc.chemie.tu-darmstadt.de/vrml/p53dna/
  12. A. W. Hewat. Details of this group's work can be seen at http://www.ill.fr/dif/3D_crystals.html
  13. O. Casher, C. Leach, C. S. Page and H. S. Rzepa, Theochem, 1996, 368, 49.
  14. For details of the Cambridge Crystallographic Database and the IsoStar system, see I. J. Bruno, J. C. Cole, J. P. M. Lommerse, R. S. Rowland, R. Taylor and M. L. Verdonk, Comp. Aided Mol. Design., 1997, 11, 525-537
  15. H. S. Rzepa, C. Leach and O. Casher, Proc. 1st Electronic Comp. Chem. Conference (ECCC-1), CD ROM Version, ARInternet Corp., Landover, Md., 1995. See also R. J. Abraham, E. J. Chambers and W. A. Thomas, J. Chem. Soc, Perkin Trans. 2, 1993, 1061.
  16. C. Conesa and H. S. Rzepa, Perkin Trans 2, 1998, 857. See http://www.rsc.org/is/journals/clic_rsc/p08165/p08165_1.htm. See also http://www.pc.chemie.tu-darmstadt.de/vrml/vib/ for examples of normal mode animation using VRML.
  17. D. James, B. J. Whitaker, C. Hildyard, H. S. Rzepa, O. Casher, J. M. Goodman, D. Riddick and P. Murray-Rust, New. Rev. Information Networking, 1996, 61; H. S. Rzepa, B. J. Whitaker, J. Goodman, O. Casher, D. Riddick, J. Griffiths, D. James, C. Hildyard, Abs. Papers Am. Chem. Soc, 1997, 214, pp.45-COMP. For examples of articles enhanced as part of the CLIC project, see http://www.rsc.org/is/journals/current/chemcomm/cccenha.htm, and the following articles in particular; http://www.rsc.org/is/journals/clic_rsc/csr398/csr398.htm and http://www.rsc.org/is/journals/clic_rsc/c04140_0.htm
  18. These models are only available via an on-line version of this article (http://www.ch.ic.ac.uk/rzepa/vrml/).
  19. The Chime plug-in from MDL Information Systems (http://www.mdli.com/chemscape/chime/) is another excellent model display tool, although lacking the generality of application of VRML.
  20. O. Casher, unpublished work.