From Protein Structure to Function with Bioinformatics.pdf

252 R.A. LaskowskiConsequently, many structures started to emerge of proteins of unknown function.Indeed, about a third of the SG structural models are of proteins of unknownor uncertain function. This rather limits their usefulness. No longer do the modelsexplain how the protein’s function is achieved as the protein’s function is not infact known.Surely, though, the structure, once known, should reveal all? After all, the historyof structural biology shows that 3D structure explains function. Virtuallyevery structure that had previously been solved had helped explain some biologicalor biochemical process. So, given the structure – hey presto – out should popthe function.Sadly, few things in life – or in bioinformatics – are that simple. The structuremay explain a function, but only if you know the function already. Despite theavailability of the many, diverse methods discussed earlier in this book, it is surprisinglydifficult to determine the function from the structure alone.10.1.1 The Problem of Predicting Function from StructureWhy is this so? Firstly, if one has a protein of unknown function it means that, notonly is there no experimental information about its function, but also that the standardsequence methods for functional annotation have failed. These methods, particularlythe various profile methods such as the Hidden Markov Model (HMM)methods, have become quite sophisticated in recent years and can detect similarityof function at quite low levels of sequence identity. So if these methods have failedwe really are relying on the 3D structure alone.The structure can provide clues to function at various levels and in varyingdegrees of reliability as the preceding chapters have described. Chapter 6 showedhow, at the global level, a protein’s fold can very often give a clue to its functionas some folds are strongly associated with certain functions. So the first step inidentifying function from structure is invariably to find a protein of known functionwith a similar fold. There are a large number of fold comparison servers onthe web that will do this, and these have been compared in several reviews (Sierkand Pearson 2004; Novotny et al. 2004; Carugo 2006). However, you need to bearin mind that a similarity of fold does not necessarily imply a similarity of function.For example, the so-called superfolds (Orengo et al. 1994; see also Chapter 6),such as the TIM-barrel family, can support large numbers of different functions(Nagano et al. 1999; Anantharaman et al. 2003). And, if the protein has a novelfold – a successful outcome in the eyes of some SG consortia – there will be nofold match at all.More locally, the surface of the protein, particularly its clefts and pockets, canhold important clues to function (Chapter 7) as can specific local arrangements ofresidues, such as those involved in catalysis, DNA recognition, etc. (Chapter 8). Soyou may be able to identify, say, a possible ATP-binding site. This would be animportant clue to function, but not the full story.