BioBrowser – Visualization of and Access to Macro-Molecular … · 2007-02-02 · BioBrowser – Visualization of and Access to Macro-Molecular Structures L.Oﬀen2 and D. Fellner1

BioBrowser – Visualization of and Access toMacro-Molecular Structures

L.Offen2 and D. Fellner1

1 Fraunhofer Institute of Computer Graphics,[email protected]

2 Institute for Computer Graphics and Knowledge Visualization,TU Graz, [email protected]

Summary. Based on the results of an interdisciplinary research project the paperaddresses the embedding of knowledge about the function of different parts/structuresof a macro molecule (protein, DNA, RNA) directly into the 3D model of this mole-cule. This way, the 3D visualization also becomes an important user interface com-ponent to accessing domain-specific knowledge – similar to a web browser enablingits users to access various kinds of information.

In the prototype implementation, named ’BioBrowser’, various information re-lated the bio-research is managed by a database using a fine-grain access control,which also supports restricting the access to parts of the material based on the userprivileges. The database is supplied by a SOAP web service so that it is possible(after identifying yourself by a login procedure of course) to query, to change orto add some information remotely by just using the 3D model of the molecule. Toannotate parts of the molecule, to change annotations, or to annotate annotationssubstructures are selected either by an easy query language or by just picking themin 3D model with the mouse.

1 Introduction

In the field of structural biology, especially when talking about drug design,the three dimensional model of an protein becomes more and more important.The reason is that the three dimensional appearance is very closely linked tothe folding or tertiary structure of a protein. This foldings is vital for thefunction of a protein, because it defines the so called active sites, which arethe parts with which other molecules can interact.

There are some approaches to predict the folding just by the sequence ofamino acids – the so called primary structure – as in [KHM+03], but thisprediction is difficult to compute and does not result in the correct foldingin all cases. So the main source for structure determinated proteins are the

2 L.Offen and D. Fellner

NMR spectroscopy and x-ray diffraction. The models of almost all of theseprotein can be found in the RCBS protein data bank [SLJ+98]. Additionalinformation to these proteins are stored in many other databases spread overthe net. These are for example the UniProt database [ABW+04], which wascreated by combining the Swiss-Prot- [BBF+05], TrEMBL- [BBA+03] andPIR- [HHSW04] databases, or the GenomeNet (http://www.genome.jp).

There are two main classes of programs for accessing the available in-formation. On the one hand there are many tools for visualizing the threedimensional model of a protein. On the other hand there are tools for thedatabase access.

Programs belonging to the first class are for example RasMol [SMW95,Ber00], an elderly program and therefore the visualization quality does notmeet todays standard, but it is still under maintenance and very wide spread.The open source project JMol [JMo] uses the java-applet mechanism to en-sure platform independence, but this results in a slower visualization. Py-Mol [DeL02] is a python based tool, which is capable of editing proteins aswell, and gOpenMol [Laa92, BLL97] is manly an graphical interface for a setof programs from OpenMol. There are many other programs like Chimera[PGH+04], Cn3D [Hog97], or FPV [CWWS03] to name just a few of them.

All of them either exhibit a significant drop in rendering performancewhen handling very large proteins or have limited visualization styles for, e.g.,ribbon structures [CB86, Ric81], molecular surfaces [Con83, LR71] or evenspacefill. Another point they have in common is the lack of some annotationpossibilities. For example to create a link to a database like the ones mentionedabove or to include additional information not yet ready for publication.

The second class of tools include the web interfaces of the different data-bases to access them and tools like BLAST to query the databases for sim-ilarity in the primary structure. But to our knowledge there is no programcombining these two aspects in the way to use the three dimensional modelof the protein as an interactive access tool for the databases. Speaking in theterms of the digital library: The 3d-model of the protein becomes the centraldocument in the daily work. This integration of the available information intothe model was one main goal of the interdisciplinary ”‘BioBrowser”’ project.

2 BioBrowser

The first challenge in this project was to provide a core application, whichprovides a plugin interface such that almost all other functionality could beimplemented as such a plugin [HOF05]. Afterwards the main visualizationstyles, which are used in the community of structural biologists, have to berealized in such a way, that they provide a quality as high as possible combinedwith an interactive frame rate. This is essential for using the 3d model as theinteractive access tool for the databases. Figures 2(a) to 2(e) show the commonvisualization styles as screen shots from the BioBrowser.

BioBrowser – Visualization of and Access to Macro-Molecular Structures 3

2.1 The Visualization Part of the BioBrowser

To gain the quality shown in the figures very different approaches from com-puter graphics were involved. Some of them use functionality which is onlyavailable on recent graphics hardware, such as pixel shaders. Fallback solu-tions are available if the program is run on older hardware, which usuallyinvolve a minor loss in display quality.

Ball and Stick, Sticks, Spacefill

All these styles use different combinations of spheres and cylinders of differentsize. To reach the quality shown in figures 2(a),2(b) and 2(c) extensive useof the vertex and fragment shader functionality of modern graphics boardsis made. Therefore the sphere and the cylinder are parameterized, e.g. thesphere by position and radius (a 4d vector) and the cylinder by a position,an axis, a radius and a length (two 4d vectors). These simple data is thentransfered to the graphics board, where a ray casting algorithm, implementedas a fragment shader, calculates the individual intersection points and normalson a per pixel level [KE04]. As fallback solutions billboarding, depth sprites,and multi resolution meshes are provided [HOF05].

Ribbons

The ribbon structure shown in figure 2(d) is realized as a combined BRep[Hav02]. This means that only a coarse base mesh is computed from thepositions of the supporting atoms – the Cα atoms – which is adaptively refinedin such a way that the resulting surface seems to be smooth independent fromthe viewpoint and passes exactly through these Cα atoms [HOF04].

Molecular Surfaces

For a fast calculation and visualization of the molecular surfaces an algorithmbased on the reduced surfaces [SO96] is combined with an adaptive tessellationof the resulting surface, which is based on second order Bezier patches. A paperdescribing this in detail is in preparation. For a brief overview have a look at[HOF05].

2.2 Selections

We can differentiate two different classes of selections. One class manageswhich parts of the molecule should be rendered in which style. Since we use aplugin system for the visualization styles, each plugin must manage this classon its own. This is because only the plugin knows which substructures arethe smallest one it can visualize separately. The other class is the selection


Atoms

Residues

Secondary Structures

A0 A1 Am

R0 R1 Rl

C0 C1

M0

Mol

Mi

Cj

SkS1S0

sel sel sel

sel sel sel

sel sel sel

sel

sel

sel

sel sel sel

Molecule

Chains

Models

Fig. 1. A selection tree (on the left) is a tree with the same layout as the moleculetree. It has links to the molecule tree and states if an item is selected or not.

which is currently manipulated, e.g. changing color, annotating, made currentselection for a visualization style, and so on. This class is managed by the coresystem. The core system provides classes for a so called selection tree (figure1). This is manly a tree of the same layout than the molecule tree with linksto the associated notes in the molecule tree. It provides fast functions for:

• Merging two selections• Subtracting one selection from another• Intersecting two selections• Inverting one selection• Iterating through a selection

So every plugin has its own selection tree, which is used for displaying purpose,and the core system holds a selection tree for manipulation issues. The latterone can be manipulated by the user by simple picking operations in the 3dmodel or by using a query language with a simple xml-syntax. The pickingoperation provide functionality for setting, adding, or removing substructuresto the current selection. For example when an atom is picked by the mouseany parent structure of this atom could be selected for the operation, e.g. theamino acid the atom belongs to. Such an interactive selection process is shownin figure 3.

For more complex selections, like selecting all oxygen atoms in a chain, asimple query language is included. The syntax of this language is xml-basedand stated in figure 1. The ”!” is interpreted as a not equal. For example toselect all atoms but oxygen the statement would be:

<item atom-element="!O"/>

To select the backbone of a chain without the nitrogen atoms use:


SELECTION: [<list>ITEM*</list]|ITEM

ITEM: <item [MOL] [MODEL] [CHAIN] [SSTRUCT] [RES] [ATOM]

[CHILDREN]/>

MOL: molecule-id=ID

MODEL: moleculemodel-id=ID

CHAIN: chain-id=ID

SSTRUCT: [secs-id=ID] [secs-type=TYPE]

RES: [res-id=ID] [res-bh=BHID] [res-name=RESNAME]

ATOM: [atom-id=ID] [atom-element=ELEMENT] [atom-name=ATOMNAME]

[atom-atomicnumber=NUMBER] [atom-function=FUNC]

CHILDREN: withChildren="yes"|"no"

ID: STRING|!STRING % id string

TYPE: TYPENAME|!TYPENAME

TYPENAME: "helix"|"sheet"|"turn"|"none"

BHID: BROOKHAVEN|!BROOKHAVEN

BROOKHAVEN:"ALA"|"ARG"|"ASP"|"ASN"|"CYS"|"GLU"|"GLN"|"GLY"|

"HIS"|"ILE"|"LEU"|"LYS"|"MET"|"PHE"|"PRO"|"SER"|

"THR"|"TRP"|"TYR"|"VAL"

RESNAME: STRING|!STRING % name of amino acid

ELEMENT: ELEM|!ELEM

ELEM: "C"|"N"|"O"|...

ATOMNAME: STRING|!STRING % pdb-columns 13-16

FUNC: FUNCTION|!FUNCTION

NUMBER: INT|!INT

FUNCTION: "M"|"E"|"S"|"B"|"3"

Table 1. The syntax of the query-language

<item atom-type="M" atom-element="!N" chain-id="A"/>


(a) Ball and Stick (b) Sticks (c) Spacefill

(d) Ribbons (e) Surface

Fig. 2. The common visualization styles (screen shots from the ”BioBrowser”). Inthe lower corner of each style a close up view is shown.

(a) One secondary struc-ture is already selected.Another is added by right-clicking on the moleculeand select add from thecontext-menu

(b) The result of the selec-tion

(c) From this selection oneresidue is removed. Againby right-clicking and se-lecting remove from thecontext menu

Fig. 3. Interactive selection process based on mouse input.


3 Annotations

An Annotation to a protein is some extra information associated with anysubstructure of this protein or – as an annotation to an annotation – to anexisting annotation. Nowadays these annotations are for example part of the”unstructured” part of the pdb-files:

REMARK 1 REFERENCE 1REMARK 1 AUTH L.W.GUDDAT,J.C.BARDWELL,T.ZANDER,J.L.MARTINREMARK 1 TITL THE UNCHARGED SURFACE FEATURES SURROUNDING THEREMARK 1 TITL 2 ACTIVE SITE OF ESCHERICHIA COLI DSBA ARE CONSERVEDREMARK 1 TITL 3 AND ARE IMPLICATED IN PEPTIDE BINDINGREMARK 1 REF PROTEIN SCI. V. 6 1148 1997REMARK 1 REFN ASTM PRCIEI US ISSN 0961-8368 0795

This annotation are difficult to parse for a computer and are normally in-terpreted as free text, resulting in imprecise search results. Therefore a morestructured way like the xml-structure, as shown in the following example,should be preferred to save annotations:

<references><reference id="1"><title value="The Uncharged Surface Features Surrounding the Active

Site of Escherichia Coli DSBA are Conserved and areImplicated in Peptide Binding"/>

<authors><author firstname="L.W." lastname="Guddat"/><author firstname="J.C." lastname="Bardwell"/><author firstname="T." lastname="Zander"/><author firstname="J.L." lastname="Martin"/>

</authors><astm_code value="PRCIEI"/><ccdc_code value="0795"/><journal><title value="Protein Sci."/><volume value="6"/><year value="1997"/><pages first="1148" last="1148"/><issn value="0961-8368"/><publisher>

<address></address><country value="US"/>

</publisher></journal>

</reference></references>


The advantages of this approach are the easier parsing, and the more efficientway of searching since it could be specified, if search searching term is thename of an author or part of the title. To handle the different annotationtypes the ”BioBrowser” has a database layout (see section 3.1) which canhandle both types of annotations – free text as well as xml.

In the ”BioBrowser” framework making annotations is as easy as makingthe selection to annotate. By selecting ”Annotate...” from the context menuthe annotation dialog (figure 4(a)) would be shown. Existing annotations are

(a) Selecting ”‘Annotate...”’ from thecontext menu, brings up the annotationdialog

(b) Annotations of annotations can bemade by right-clicking on an existingannotation. This will be shown by aneedle on the needle (dark arrow)

visualized as a kind of needles sticked into the protein. When such a needleis selected for an annotation this annotation will be annotated. Thereby acomplete tree of annotations is possible.

3.1 Database

All the annotations are saved into a large database. The layout of this databaseis shown in figure 4. The Layout can be split into five main Parts:

• User management (very dark grey block)• XML Mapping (dark grey block on the left)• Selection Mapping (light grey block at the top)• Annotations (grey block on the right)• Access management (grey blocks in the user management block)

User Management

The user management is inspired by the file access management of the unix filesystem. It differentiates between groups, e.g. all, institution, working group,


Fig. 4. Database layout

and the users themself. Each group respectively user has some identifying dataassigned to, such as name or description and each user can be assigned to asmany groups as needed. For each user some more rights can be specified onthis level: the right to change user or group data and the administrative accessto the database. For identifying an user by an login mechanism an encryptedpassword for each user is stored as well.

XML Part

The xml-part maps the tree structure of a xml-file into the flat structure of arelational database. This is done by saving the parent id and the root id of thetree into each node (xmlentry). The entry also saves the used token and hasa mapping to the different attributes this node has. The token is not saveddirectly, but uses only a link to another table collecting all tokens. Therebythe complete xml-file can be reconstructed due SQL-Queries.


Selection Mapping

A selection is saved into the database by splitting it up its selection tree intothe different selected items, e.g. chains, residues. Each such item is insertedinto the database using a hash value for faster access. A whole selection hasa mapping to all substructures it is composed of and also uses a hash value,computed from the hash values of these items. By this a query for a spe-cific selection can be done very fast. First search for all selections having thesame hash value as the desired one and afterwards check the resulting onesif the items they are composed of match the ones the desired one is build of.Depending on the used hash functions the first query will result in only oneresulting selection, which have to be approved by matching it selected itemswith the desired ones.

Annotations and Access management

Through the access management of the database a fine access control canbe assigned to every annotation. The owner or creator of an annotation canspecify who can read or change the annotation by assigning the desired rightsto the users and/or groups. Afterwards only the users matching the assignedrights have the possibility to access the annotation with the rights specifiedby the owner.

An annotation consists of a link to the root node of the assigned xml-structure and additionally may contain a plain text version as a note. If theannotation is annotating another annotation the id of it will be added asparent id to build the tree structure of the annotations. As another possibilityto access the annotations every annotation can be feed with keywords.

3.2 Database Access

To access the database structure described in the last section the frameworksupports two different approaches. On the one hand it has an ODBC interface,which can connect to local databases. On the other hand a web service basedon the SOAP protocol [W3C] including the needed server has been developed.These two approaches are explained in more detail below. But first the mainscenarios for accessing the database are shortly presented:

Logging into the database This is the first step, which must be done whenusing the database. Here the given user and password are checked againstthe data in the database. If this check is positive the user has access tothe database.

Administrative Access If the current user has the administrator flag set,all SQL instructions are forwarded to the database. This is only necessaryfor low level access and there should be at most one user with this flagset.


Changing the user management data When the user has the right tochange the user data the framework will allow the user to add new groupsor users respectively change the old ones.

Querying all annotations for a given selection This setting is mainlyused when inserting the known annotations into the 3d model of the pro-tein. When this visualization module is activated the database is queriedevery some frames for the known annotations of the current moleculewhich the current user has the right to read of course, so changes fromother users will incorporated after only some frames. Then the selection ofeach annotation is parsed and the annotation is placed at an appropriateposition in the model.

Updating annotations Using this the user can add new annotations forany selection or to any old annotation. He can also change the content ofan old annotation if he has write access to this annotation. The data thatcan be changed of course includes the access rights of these annotations.So for example an annotation could be opened for the public, when theresults are published elsewhere.

Searching for annotations having a given property This query resultsin a list of annotations having the given property, which can be a keyword,a phrase in the plain text section, or a phrase in some specific section ofthe xml structure of the annotation.

SOAP Web service

To share the database with remote users a so called SOAP web service, basedon a xml protocol, has been implemented. Thereby a client sends xml format-ted queries to a server. This server interprets the queries, processes them andgenerates the answer, which is send back to the client.

The protocol for the allowed queries is defined in the web service descrip-tion language (wsdl) and can be retrieved from the server. So the access tothe database is not restricted to the ”BioBrowser” framework but every clientimplementing this protocol could access the database. Since the server inter-prets every query, it can check whether the current user is allowed do executethis query or not. When the user has the required rights, the answer is sendback to the querying application, otherwise an appropriate error message issent.

ODBC Interface

When using the ODBC protocol, which the operating system must provide,the framework is responsible for the access control. Whenever a query to thedatabase is submitted, the result is checked against the rights of the user. Ifthe user currently logged into the system hasn’t the right to read the result isdiscarded. Otherwise it is presented to the user. If he has the right to change


it any changes will be submitted to the database. When the user has only therights to read all changes are discarded.

The problem with the direct ODBC protocol is, that the framework couldbe circumvent through this protocol. This means that every user connected tothe database through this protocol can sent arbitrary low level SQL queries.By this the use of this interface is really restricted to the access to local data-bases or databases protected from unrestricted access by a firewall. For remotedatabases the web service should be used, since then all SQL instructions areparsed by the server and discarded when the current user does not have theappropriate rights.

4 Results

This paper demonstrate how to integrate the knowledge about biomolecularstructures into the three dimensional model of the structure. This is exampli-fied with the prototype implementation of the ”BioBrowser”. For an usableintegration two different problems had to be solved:

First of all the visualization of the three dimensional model must be fastenough and of high quality so that the three dimensional model of the mole-cule can be introduced as the main document for the daily research work.This part has been overcome by consequent use of the available hardwareand modern computer graphics algorithms. The second problem is connectedto the integration of the available knowledge into this model. Therefore the”BioBrowser” uses an annotation mechanism, which connects the knowledgeabout parts of the molecule directly to the corresponding parts of the model.This annotation are saved into a database which can be queried locally by anODBC interface or remotely by a SOAP web service. To restrict the acces-sibility of certain annotations, a fine-grain access control is included into thedatabase.

References

[ABW+04] R. Apweiler, A. Bairoch, C.H. Wu, W.C. Barker, B. Boeckmann,S. Ferro, E. Gasteiger, H. Huang, R. Lopez, M. Magrane, M.J. Mar-tin, D.A. Natale, C. O’Donovan, N. Redaschi, and L.S. Yeh. Uniprot:the universal protein knowledgebase. Nucleic Acids Res., 32:115–119,2004. http://www.uniprot.org.

[BBA+03] B. Boeckmann, A. Bairoch, R. Apweiler, M.-C. Blatter, A. Estreicher,E. Gasteiger, M.J. Martin, K. Michoud, C. O’Donovan, I. Phan, S. Pil-bout, and M. Schneider. The swiss-prot protein knowledgebase andits supplement trembl in 2003. Nucleic Acids Res., 31:365–370, 2003.http://www.expasy.ch/cgi-bin/sprot-search-ful.

[BBF+05] B. Boeckmann, M.-C. Blatter, L. Famiglietti, U. Hinz, L. Lane,B. Roechert, and A. Bairoch. Protein variety and functional diversity:


Swiss-prot annotation in its biological context. Comptes Rendus Biolo-gies, 328:882–899, 2005.

[Ber00] H. J. Bernstein. Recent changes to rasmol, recombining the variants.Trends in Biochemical Sciences, 25:453–455, 2000. http://www.rasmol.org/.

[BLL97] D.L. Bergmann, L. Laaksonen, and A. Laaksonen. Visualization of sol-vation structures in liquid mixtures. J Mol Graph Model, 15:301–306,1997.

[CB86] M. Carson and C.E. Bugg. Algorithm for Ribbon Models of Proteins.J.Mol.Graphics, pages 121–122, 1986.

[Con83] M.L. Connolly. Solvent-accessible surfaces of proteins and nucleic acid.Science, 221:709–713, 1983.

[CWWS03] Tolga Can, Yujun Wang, Yuan-Fang Wang, and Jianwen Su. FPV:fast protein visualization using Java 3DTM. In Proceedings of the 2003ACM symposium on Applied computing, pages 88–95. ACM Press, 2003.http://www.ceng.metu.edu.tr/∼tcan/fpv/.

[DeL02] W. L. DeLano. The PyMOL Molecular Graphics System. DeLano Sci-entific, San Carlos, CA, USA, 2002. http://www.pymol.org.

[Hav02] S. Havemann. Interactive Rendering of Catmull/Clark Surfaces withCrease Edges. The Visual Computer, 18:286–298, 2002.

[HHSW04] H. Huang, Z.Z. Hu, B.E. Suzek, and C.H. Wu. The pir integrated proteindatabases and data retrieval system. Data Science, 3:163–174, 2004.

[HOF04] Andreas Halm, Lars Offen, and Dieter Fellner. Visualization of ComplexMolecular Ribbon Structures at Interactive Rates. In Proceedings of theInformation Visualisation, Eighth International Conference on (IV’04),pages 737–744. IEEE Computer Society, 2004.

[HOF05] Andreas Halm, Lars Offen, and Dieter Fellner. BioBrowser: A Frame-work for Fast Protein Visualization. In Ken Brodlie, David Duke, andKen Joy, editors, Eurographics / IEEE VGTC Symposium on Visualiza-tion, pages 287–294, Leeds, United Kingdom, 2005. Eurographics Asso-ciation.

[Hog97] C. W.V. Hogue. Cn3D: a new generation of three-dimensional molecularstructure viewer. Trends in Biochemical Sciences, 22:314–316, 1997.ftp://ftp.ncbi.nih.gov/cn3d/.

[JMo] Jmol. http://www.jmol.org.[KE04] T. Klein and T. Ertl. Illustrating Magnetic Field Lines using a Discrete

Particle Model. In Proceedings of the Workshop on Vision, Modelling,and Visualization 2004 (VMV ’04), pages 387–394, 2004.

[KHM+03] Oliver Kreylos, Bernd Hamann, Nelson L. Max, Silvia N. Crivelli, andE. Wes Bethel. Interactive Protein Manipulation. In Proceedings of the14th IEEE Visualization Conference 2003, 2003.

[Laa92] L. Laaksonen. A graphics program for the analysis and display of mole-cular dynamics trajectories. J Mol Graph, 10:33–34, 1992.

[LR71] B. Lee and F. M. Richards. The interpretation of protein structures:Estimation of static accessibility. J. Mol. Biol., 55:379–400, 1971.

[PGH+04] E. F. Pettersen, T. D. Goddard, C. C. Huang, G. S. Couch, D. M. Green-blatt, E. C. Meng, and T. E. Ferrin. UCSF Chimera - A visualizationsystem for exploratory research and analysis. Journal of ComputationalChemistry, 25:1605–1612, 2004. http://www.cgl.ucsf.edu/chimera/.


[Ric81] J.S. Richardson. The anatomy and taxonomy of protein structure. Adv.Protein Chem., pages 167–339, 1981.

[SLJ+98] J. L. Sussman, D. Lin, J. Jiang, N.O. Manning, J. Prilusky, O. Rit-ter, and E.E. Abola. Protein Data Bank (PDB): database of three-dimensional structural information of biological macromolecules. ActaCrystallogr., D 54:1078–1084, 1998. http://www.pdb.org.

[SMW95] R. Sayle and E. J. Milner-White. RasMol: Biomolecular graphics for all.Trends Biochem. Sci., 20:374, 1995. http://www.rasmol.org/.

[SO96] Michel F. Sanner and Arthur J. Olson. Reduced Surface: an EfficientWay to Compute Molecular Surfaces. Biopolymers, 38:305–320, 1996.

[W3C] W3C. SOAP Version 1.2 Part 0: Primer.

Documents

BioBrowser – Visualization of and Access to Macro-Molecular … · 2007-02-02 · BioBrowser – Visualization of and Access to Macro-Molecular Structures L.Oﬀen2 and D. Fellner1