A multi-start Monte Carlo procedure restrained by surface geometry match is presented and used to dock three serine protease-peptide segment systems. For each of the ligands, one hundred of initial conformations, which meet given steric and geometric tolerances, are selected randomly. Each of the initial conformations is optimized by a local Monte Carlo procedure which makes a step to a new random step along with the routine of reduction of the radius of gyration Rg of protein-ligand complex according to a given probability. The random step is followed by a local minimization in the total conformational spaces. The radius of gyration Rg of protein- ligand complex characters the quality of geometric fit. Small Rg value corresponds to compact packing and good fit of geometry on the interface. The algorithm successfully prddicts the conformations of ligands close to the crystal structures for the three test systems in the docking of rigidity, partial flexibility and total flexibility and the runs display much better convergence properties than the procedure without the guide of geometric fit.

Key words: MONTECARLO SIMULATIONS, PROTEIN, IDENTIFICATION, SERINE, PROTEASE