Predicting immunoglobulin-like hypervariable loops

A two-stage method is developed to search the conformational space of small protein segments for low energy structures. Central features of the method are efficient procedures for generating small, eight-backbone atom, local moves in Cartesian coordinates and for introducing geometric constraints in adaptable Monte Carlo procedures. This allows natural implementation of an adaptive simulated annealing algorithm, which achieves an effective trade-off between speed and acceptance ratio. The method is applied to the calculation of various immunoglobulin loops. We also develop data base derived rules for identifying constraint condition, and show that the incorporation of an identified side-chain constraint allows a 1.2 A all-backbone atom rms deviation prediction of a 9 residue long L1 loop.