Arena3D is a highly interactive, 3D visualization system – to illustrate the value of such a system, it is clearly not optimal to use only static 2D images. In this section, we present several relatively small datasets that illustrate how Arena3D can be used to gain insight into biological networks; however we refer interested readers to http://arena3d.org
, where they can download the program together with several example datasets.
In Figure we show how Arena3D can be used to illustrate the results of a query across multiple databases. In this case, we start the query from Huntington's disease, and we show the nine most strongly associated genes listed in the String database. These genes are then shown linked to 10 proteins on the next layer, where we indicate how one gene (Huntingtin) has two variants – mutant and wild type Huntingtin. On the next layer, we have shown all known 3D structures (75) that are linked to these proteins. One of these proteins, p53, is linked to over half of these structures, indicating that a lot of 3D information is available for this protein. In contrast, other proteins, including Huntingtin, are not associated with any known 3D structure. The graph shows one 'hidden' feature that was of interest and was not known to us before creating this view: two of the proteins associated with Huntingtin – p53 and CREB-BP – occur in the same PDB file. This means that the 3D structure of the complex of these two proteins is known.
In Figure , we show a slightly more complex network – this time starting from nine polyQ-related diseases, including Huntingtin. Each of these diseases is associated with a polyQ protein, and from the String database we have found 66 proteins associated with these proteins.
On the second layer we use affinity propagation to cluster these proteins. The graph reveals several interesting and hidden features, namely that six proteins associated with Huntingtin are also associated with other polyQ disease proteins: they are: TBP (Sca17), Atn1 (DRPLA), Atxn1 (Sca1), Atxn7 (Sca7), CACNA1A (Sca6), and PQBP1 (Sca1).
On the final layer we have mapped all 151 domains present in the 66 proteins, and we have highlighted eight domains that are present in these six proteins shared with Huntingtin and other polyQ diseases. This highlighting reveals another interesting hidden feature: two domains are common in two diseases: WW is present in both ITCH (Sca7) and PQBP1 (Sca1), and the Atrophin domain is present in both Atn1 (DRPLA), and RERE (Sca7).
In Figure , we show the nine proteins associated with Huntingtin mapped onto the sub-cellular location hierarchy in GO. The view is generated by finding all GO location terms in the corresponding Uniprot [33
] files, finding the related GO hierarchy using POSOC [19
], and using the hierarchical and circular layout methods in Arena3D. As an example for how a user would navigate through such a complex view, we have coloured terms in the GO hierarchy that are directly related to the protein rasa1 – from this we see that this protein is located in protein complexes, in the apoptosome, the telomeric regions of the chromosome, and in dendrites.
Finally, Figure shows a 'full blown' illustration of how a user could work with Arena3D in practice to explore complex, large datasets. The scenario illustrated shows again the nine proteins associated with Huntingtin. In this case the rasa1 protein is selected, and its GO terms are shown together with the GO hierarchy, this time using a planar layout. For illustration, we have shown the proteins mapped onto a group of 80 related chemicals clustered via affinity propagation using Tanimoto similarity. The chemicals are also shown clustered using a tree method. In this case, to simplify the graph, connections between the tree layer and others layers have been hidden by Arena3D.