computer modeling in molecular biology.pdf

22 Tim .l I! Hubbard and Arthur M. Leskthe core as a function of sequence divergence. In pairs of distantly related proteinsthe size of the cores can vary: In some cases the fraction of residues in the core remainshigh, in others it can drop to below 50% of the structure.2.3.2 TechniquesA general outline of the steps involved in model building by homology is as follows:1. The sequence of unknown structure is aligned to the sequence(s) of known structureand the sequence of unknown structure is divided into SCR’s and SVR’s:regions where the alignment has sufficient sequence conservation to be conservedstructurally (alignable) are defined as SCR’s (Structurally Conserved Regions).The remaining regions (nonalignable - including but not restricted to loopregions) are defined as SVR’s (Structurally Variable Regions).2. The main chain conformation and spatial relationship of the SCR’s are takenfrom the coordinates of the known structure to which they were aligned. Conformationsare generated for each SVR in the sequence, with correct endpointgeometry and length, which do not clash sterically with the rest of the structure,either using a database search method [44] or any alternative approach. Thiscreates a complete continuous main chain model.3. Side chains are built onto the main chain model and their conformations optimised.2.3.2.1 Alignment and Division into SCR’s and SVR’sDiscovering a relationship between an input sequence and a structure does notnecessarily give a full or accurate alignment. Frequently more sensitive alignmenttechniques not designed for fold recognition can give a more accurate alignmentonce the sequences to align have been identified. It is important to keep the lessonsfrom evolution in mind (Section 2.3.1): There will be regions where there are severalpossible alternative alignments and there will be regions that cannot be aligned(nonalignable) because they are structurally different.The first of these problems can be tackled by using as much information as possible(multiple sequence alignments for both known and unknown) and by exploringsignificant alternative sub-optimal alignments [45, 461, and if necessary, by buildingmultiple structural models using alternative alignments.The second problem is to distinguish the regions in which the input sequence hasthe same fold as the model (SCR’s) and where it is different (SVR’s). Clearly wherethere are insertions and deletions the chain trace of the model must be different;