computer modeling in molecular biology.pdf

30 Tim J.l? Hubbard and Arthur M. Leskwidespread. Such methods make use of multiple sequence information, looking forconsistency between the predictions for different sequences. It should be noted thateven given a correct secondary structure assignment, it is very difficult to determinehow the units fit together in three dimensions. [77]. A table of aligned sequences maywell contain derivable information about the 3-D structure of a protein but attemptsto recover it have so far met with no more than sporadic success [12]. However, usefuldeductions about the most likely folded structure can be made in a systematic wayfrom a combination of analysis of SSP results and the conservation patternsobserved in a multiple sequence alignment. Successful predictions using such an approachhave been made for the annexin [78] and Src homology 2 (SH2) proteinfamilies [79].2.4.2 A Lone Sequence or a Designed Sequence:no Multiple Sequence, no Known RelativesThis situation is the most unfavourable for model building; as one has no way ofapplying known sequence or structure information. In effect the problem can onlybe handled by a priori methods, Even secondary structure prediction is inaccuratefor single sequences and therefore the likelihood of building a correct three-dimensionalmodel is small.If there is any suspicion (perhaps on functional grounds) that a natural sequencehas a certain fold, or in the case of a designed sequence, built to fold in a particularway, the situation is slightly better since it is possible to test the likelihood that a sequencecan match a particular fold. Methods for doing this include checking polarity1801; packing quality and residue-residue contact frequencies [81] ; various freeenergy functions incorporating solvation effects [82, 831, hydration and heat stabilityeffects and more recently using threading techniques to establish if the sequence iscompatible with the fold [MI.The disadvantages of these methods are that (1) most provide only an assessmentof the structure as a whole rather than of local regions (models are frequently onlypartially right e. g. [86]) (2) even at this level they are inaccurate, i. e. some experimentallydetermined structures are classified as incorrect whereas some misfolded modelsare classed as correct in blind tests and (3) that the results are essentially dependenton the quality of the model rather than the correctness of its fold. Moreover, thesetests only look at the final state and do not assess if the sequence is compatible withany pathway to that state. For natural sequences it can be assumed that folding toa compact state can be achieved but this is more likely to fail to be the case fordesigned sequences. Current experimental [87] and theoretical work [88] on thefolding pathways of proteins suggest that there are clear folding initiation sites