Several problems in protein biochemistry, such as functional analysis, transition state analysis, structure refinement, and flexible-receptor molecular docking, can be reduced to large-scale sampling of the low-energy part of a protein's conformation space. In this work, we present a method for sampling the local conformation space of a whole protein, with thousands of degrees of freedom, starting from a known, or at least low-energy, input structure. The method combines dimensionality reduction using principal component analysis with a sampling algorithm inspired by the Expansive Spaces robotic path planner, which was designed to rapidly cover a large, high-dimensional robotic configuration space. Results are presented for tests of the search on two whole-protein models: HIV-1 protease, with 198 amino acid residues (3120 atoms), and FK506 binding protein (FKBP), with 107 residues (1663 atoms). Experiments were performed with several variants of the search algorithm, starting from documented structures determined by x-ray crystallography and NMR. Each variant was evaluated on its ability to produce a set of diverse, low-energy structures of the input protein.
展开▼
