Existing Methods and Limitations Knowledge-Based B-Factor Restraints for Knowledge-Based B-Factor Restraints for

Introduction

The principal problem with macromolecular refinement is the lack of diffraction data in relation to the number of parameters in the model. In many cases the diffraction from a crystal cannot be measured beyond 3Å or so because the intensities become too weak. In a typical case, assuming the solvent content is 50%, there will only be 4 to 5 reflections per ordered atom. The standard form of the structural models used in refinement contains four parameters per atom. The diffraction data set does not contain sufficient information to define all the parameters of the model.

To supplement the diffraction data one generally adds as observations relationships between the atoms of the model derived from higher resolution models. In the restrained refinement method currently in general use (PROLSQ [Hendrickson & Konnert, 1980] XPLOR[Brunger et al., 1987], and TNT[Tronrud et al., 1987]) the positions of the atoms are restrained to conform to the expected bond lengths, bond angles, and other positional information. Considerable attention has been paid to the derivation of precise libraries of stereochemical restraints using small molecule models [Kennard, 1968], [Levitt, 1974], [Vijayan, 1976], [Kennard & Taylor, 1982], [Saenger, 1983], [Allen et al., 1987], [Engh & Huber, 1991].

However, very little has been done to identify restraints which could be applied to the B-factors of a macromolecular model. This paper describes an attempt to restrain B-factors using a library of restraints derived from a small collection of well refined protein models. The proposed method is compared to the method currently in use.