computer modeling in molecular biology.pdf

2 Modelling Protein Structures 271. Align the sequence of the VL and VH chains of the target immunoglobulin withthe sequences of the corresponding domains in the known immunoglobulin structures.2. For each domain (VL and VH), select a parent domain from among the correspondingdomains of known structure. The percent residue identity with thetarget domain is usually in the range 45 070 and 85 070.3. If the selected parent structures for VL and VH domains come from different immunoglobulins,pack them together by a least-squares fit of the main chain atomsof residues conserved in the VLVH interface.4, Identify the canonical structure of each loop by checking the sequence for theparticular sets of residues that form the signature of each canonical structure. H3is a special case, far more variable in length, sequence and structure; and mustbe modelled by other methods.5. Graft a loop from a known immunoglobulin structure - preferably the domainfrom which the framework was built - into the framework model.6. If a canonical structure for any loop cannot be identified, the loop must bemodelled by other means.7. To build the side chains: At sites where the parent structure and the model havethe same residue, retain the conformation of the parent structure. If the side chainis different, take its conformation, if possible, from an immunoglobulin havingthe same residue in the corresponding position; within hypervariable loops, takethe side chain conformation only from a loop with the same canonical structure.8. Subject the model to limited energy refinement, only to tidy up the stereochemistry.How good a model can be expected from this procedure, assuming the hypothesisthat for the three loops of the light chain and for the first two of the heavy chaina canonical structure present in the data base can be identified?The first of several “blind” tests was made on the antilysozyme antibody D1.3[6]. Comparison of this prediction with the best available crystal structure of D1.3has shown that all six hypervariable regions had the predicted main chain conformations.Other tests are described in Chothia et al. [66]. The general conclusion is thatif the structures used as parent structures for the two domains and the loops are highresolution, well-refined structures, one can expect the backbone of the frameworkto be correct within 1.0 A r. m. s. deviation, and the backbone of the predicted loops,not including the special case of H3, to differ by about 0.7 A r.m.s. deviation onaverage, and by no more than 1.0-1.2 A in all cases. In addition, one can expect thepositions of Ca atoms of residues in the loops to shift, relative to the frameworksof VL and VH domains, by 1.0-2.0 A typically and by up to 3 A in the worst cases.