From Protein Structure to Function with Bioinformatics.pdf

294 I.A. Cymerman et al.types, scientists often focus on seeking functions for the vast numbers of sequencescollected in the databases. Obviously, the change from the descriptive to the morepredictive applications requires the development of novel tools.The most basic information of interest regarding a particular gene or proteinis its associated function. The most common approach to function predictionstarts from the observation that proteins with similar sequences frequently havesimilar functions. Particularly with the recent increase in the number of availablesequences, related sequences can be aligned and grouped into families. If thefunction of one of the family members is known, the remaining sequences arehypothetically assigned with the same function “through inheritance”. This raisesthe question as to whether one needs a protein structure to predict protein functionalityor is sequence information sufficient. At first sight one could answer thatit depends on the sequence similarity between two compared protein sequences.It is widely accepted that sequence identity higher than about 30% is a strongindicator that the proteins may share a very similar structure that can be predictedby comparative approaches (see Chapter 3), yielding generally accurate models.Below this threshold, however, assignment of protein structure requires moresophisticated approaches (such as those outlined in Chapters 1 and 2) and is nolonger as accurate. Since function depends on structure, one could envisage thata similar dependency holds also for transfer of functional annotations. This, however,is not necessarily the case, as significant functional variation can beobserved even for proteins with very similar sequences and structures. Forinstance, functional annotations based on Gene Ontology are conserved in only80% of protein pairs even for proteins sharing 90–100% identity; this value dropsbelow 50% for proteins with less than 30% sequence identity. Some aspects offunction appear to be more conserved than others, e.g. if the enzymatic functionis considered according to the EC system, all four EC numbers tend to be conservedin nearly 100% cases above 70% pairwise sequence identity. For sequenceswith pairwise identity below 30%, the conservation of EC numbers also dropsbelow 50% (Tress et al. 2008).Conservation of function is more complex than conservation of structure,because functional overlap (e.g. identical function in two copies of one gene aftera duplication) is subjected to evolutionary rate acceleration, which howeverdepends on pre-existing functional utility of the protein encoded by the ancestralsingleton (Jordan et al. 2004). Thus, a duplication that gives rise to paralogousproteins with very similar sequences and structures either results in a loss, byinactivating mutation or deletion, of one the copies (thus reversal to the ancestralstate) or in functional divergence of one or both copies, as to reduce the overlap.On the other hand, orthologous sequences tend to retain identical function, oftendespite considerable sequence divergence. Pairwise sequence comparisons arehowever unable to distinguish between orthologous and paralogous sequences,thus arguing against its use for function annotation. There are a number of methodsthat carry out function prediction based on evolutionary analyses, and discriminationof paralogous vs orthologous sequences (e.g. FlowerPower(Krishnamurthy et al. 2007) ). However, these methods require the availability of