Study Weekend Program... UP


Refinement of severely incomplete structures with Maximum Likelihood in BUSTER-TNT

Pietro Roversi*#, Eric Blanc*&, Clemens Vonrhein*,Claus Flensburg*, Susan Lea# and Gérard Bricogne*
* Global Phasing Ltd., Sheraton House, Castle Park, Cambridge CB3 0AX,
# Laboratory of Molecular Biophysics, Biochemistry Department of Biochemistry, Oxford University, South Parks Road, Oxford OX1 3HQ,
& European Bioinformatics Institute, Wellcome Trust Genome Campus, Hinxton, Cambridge, CB10 1SD

We present a description of the BUSTER-TNT [1] package for macromolecular structural refinement and illustrate its applications to the refinement of severely incomplete structures.

Besides the traditional partial structure and bulk solvent models, BUSTER[2] has a model for the missing atoms, based on low-resolution real space distributions[3]. The BUSTER structure factor distribution in the complex plane is a 2D Gaussian centered around the calculated offset; its variance is computed from Luzzati imperfection factors. BUSTER generates the missing atoms and solvent models, accepts from TNT[4] the partial structure Fs, assembles the total structure factor distribution, performs maximum likelihood refinement of overall scaling and imperfection parameters, and passes Rice likelihood derivatives with respect to structure factor components to TNT. TNT uses Agarwal chaining to turn these derivatives into derivatives with respect to atomic parameters, merges them with the ones coming from NCS and stereochemical restraints/constraints, and shifts the atomic parameters.

Examples taken from recent real-life refinements starting from severely incomplete Molecular Replacement models are presented and discussed[5,6,7]. When the partial structure is very incomplete, inclusion of the missing atoms model: i)gives a more accurate refinement of the overall scale factors; ii) reduces the bias affecting partial structure refinement by accounting for some of the scattering from the as yet unmodelled missing atoms; iii) improves the Luzzati and sigmaA-based error model by adding a spatial definition to the source of incompleteness; iv) provides a density modification improvement of electron density in the missing atoms regions which helps structure building and completion.

[1] G. Bricogne and J.J. Irwin, Proceedings of the CCP4 Study Weekend (1996)
[2] G. Bricogne, Acta Cryst. (1993) D49, 37-60
[3] P. Roversi et al., Acta Cryst. (2000) D56(10), 1316-23.
[4] D. Tronrud et al., Acta Cryst. (1987) A43, 489-501
[5] K. Ng et al., Nat Struct Biol. (2000) 7(8), 653-7
[6] M. Hanzal-Bayer et al., EMBO J. (2002) 21(9), 2095-106
[7] P. Lukacik et al., PNAS (2003) in the press