TORSION ANGLE DYNAMICS - REDUCED VARIABLE CONFORMATIONAL SAMPLING ENHANCES CRYSTALLOGRAPHIC STRUCTURE REFINEMENT

A reduced variable conformational sampling strategy for macromolecules based on molecular dynamics in torsion angle space is evaluated using crystallographic refinement as a prototypical search problem. Bae and Haug's algorithm for constrained dynamics [Bae, D.S., Haug, E.J. A recursive formulation for constrained mechanical system dynamics. Mech. Struct. Mach. 15:359-382, 1987], originally developed for robotics, was used. Their formulation solves the equations of motion exactly for arbitrary holonomic constraints, and hence differs from commonly used approximation algorithms. It uses gradients calculated in Cartesian coordinates, and thus also differs from internal coordinate formulations. Molecular dynamics can be carried out at significantly higher temperatures due to the elimination of the high frequency bond and angle vibrations. The sampling strategy presented here combines high temperature torsion angle dynamics with repeated trajectories using different initial velocities. The best solutions can be identified by the free R value, or the R value if experimental phase information is appropriately included in the refinement. Applications to crystallographic refinement show a significantly increased radius of convergence over conventional techniques. For a test system with diffraction data to 2 Angstrom resolution, slow-cooling protocols fail to converge if the backbone atom root mean square (rms) coordinate deviation from the crystal structure is greater than 1.25 Angstrom, but torsion angle refinement can correct backbone atom rms coordinate deviations up to approximately 1.7 Angstrom. (C) 1994 Wiley-Liss, Inc.