computer modeling in molecular biology.pdf

14 Tim Ll? Hubbard and Arthur M. Lesk14 Yo of sequences could be associated with a protein of known structure using standardsequence alignment methods.For the output, the dividing line may soon become blurred if it becomes possibleto predict more than just secondary structure (1-D) when families of homologous sequencesare considered together. The Yeast chromosome I11 analysis found that 24%of sequences could be associated with an existing sequence family that had no knownstructure. As more proteins are sequenced such families are coming to have increasinglylarge numbers of members with wider sequence diversity. Since related sequencesmay all be expected to adopt the same fold, any prediction must be consistentwith each sequence in such a family. This is a considerable restriction and hasallowed significant improvements in 1-D secondary structure prediction [12- 141, thelatter method being available to anyone with access to electronic mail (send “help”to Predictprotein @ embl-heidelberg-de). Since the number of natural folds isthought to be finite and may be as small as 1000 [I51 there will come a time whenall new sequences can be associated with a known protein structure. There istherefore something of a race between various methods - fold recognition versusfold prediction - that seek to eliminate the current “unpredictable” region of sequencespace.Figure 2-1 does not include all protein modelling exercises, as it omits designedsequences. It is important to realise that even if methods to predict a structure consistentwith a large family of sequences are developed, this is not a solution of thefolding problem. The assumptions that (1) any sequence folds and (2) folds aresimilar among homologous sequences are based on evolutionary reasoning, for sequencesthat do not fold would be selected against and would not therefore beobserved by chance, and significant sequence homologies are only likely to occurthrough divergent evolution, i. e. from a single fold. Designed sequences may not foldlike the sequence to which they appear to be related and in many cases may not foldat all. In order to be able predict the structure of a designed sequence it will benecessary to predict structure from individual sequences, ignoring evolutionary relations,i.e. to solve the a priori folding problem [16].2.2 Proving a Sequence/Structural RelationshipThe first stage in any modelling project should be to compare the sequence of theprotein of interest with the contents of sequence databases. There are many sequencealignment programs available that can do this with varying speed and sensitivity. Theobjective is to find homologous sequences of known structure, but finding anyhomologous sequence is useful since it provides additional information about theprotein to be modelled.