From Protein Structure to Function with Bioinformatics.pdf

288 J.D. Watson and J.M. ThorntonPDB has its own page automatically seeded using information retrieved from OCA(Prilusky 1996) and other sources. There are a number of key difference inProteopedia compared to PDBWiki, the greatest of which is its unique linking oftext with scenes in a Jmol viewer (rather than a static image of the protein,Proteopedia provides a fully interactive Jmol view of the entry). Using a sceneauthor system, anyone editing a page can easily create scenes to highlight keyregions of a protein or restrict the view to the region being discussed in the text.This makes the Proteopedia resource more than just a set of static pages to displayinformation, instead providing an extra layer of interaction that can be used tomore easily demonstrate ideas and illustrate regions of interest. The site is alsounique in that there are no anonymous edits and the user’s full name is credited ascontributing to the page. This approach has one additional advantage: users canhave their own visible but non-editable areas in the system. This allows for thegeneration of topic-based or example pages that remain stable and can thereforebe used as a teaching tool as well as a community annotation tool.11.5 ConclusionsThere have been a great number of structures deposited by the Structural Genomicsgroups which have contributed to our treasury of protein 3D structures. However,due to the rapid data release required by these projects a large proportion of structureshave little or no functional information. The ability to predict a protein’s functionfrom sequence and structure has been something of a Holy Grail forbioinformaticians and consequently a wide variety of methods have been developedover the years. Each of these methods has its own pros and cons and there arenumerous individual case studies highlighting how one particular method can providebiological insight where others failed. No single method at this time shows a100% success rate and therefore the continued development of novel structuralanalysis and function prediction tools is essential to helping our understanding.There have been a few attempts to perform large-scale analyses of the ability topredict protein function from structure with some success but there are a few keyproblems that need to be addressed in this field. One of the greatest problems facedwhen developing function prediction methods from structure is that there is no real“gold standard” dataset to test on. As a result each method developer has had topick their own data to test on, resulting in difficulties when comparing differentmethods to one another. The second major hurdle that needs to be addressed is theproblem of comparing structure-based approaches against sequence-based methods.Although not impossible in all cases, the ability to remove information fromthe sequence databases and the patterns and profiles constructed using them is amuch more difficult problem. This only applies to “after the event” types of analysis(like that performed by Watson et al. 2007). If, instead, all function predictionswere made on the day of the structure’s release, to be stored and analysed at a future date,this problem would be overcome. This process has already been started using the