Present capabilities of BUSTER

Contents

• Overview
• Required input
• Desired output
• Model ingredients
• Refinement and completion with BUSTER
• Just phasing
• Maximum-Likelihood partial structure refinement
• Maximum-Entropy partial structure completion

Overview

Required input

• a partial structure, referred to as the fragment;

• an estimated chemical composition for the part of the structure not represented in the fragment, referred to as the missing atoms, or the rest;

• [optional] an initial indication of the non-uniformity of the missing atoms, e.g. in terms of a rough envelope for the whole molecule from which the fragment envelope can be excluded to produce the prior distribution for the (random) missing atoms; in absence of a map or a PDB model to determine this missing atoms distribution, the program can compute it on the basis of the current phases only - so this item is not strictly speaking required from the user;

• an estimate of the mean electron density of the mother liquor (or solvent) outside the molecule, and of the temperature factor for the atoms in the solvent region;

• experimental x-ray diffraction data consisting of native amplitudes and their standard deviations, possibly accompanied by experimental phase information (e.g. Hendrickson-Lattman coefficients from SHARP)

Desired output

• refined parameters for the fragment;

• a more detailed map of the distribution of the missing atoms, hopefully capable of allowing further model building beyond the initial fragment.

Model ingredients

• a PDB file and a TNT sequence file describing the current fragment; we will use the symbols PDBFRG and SEQFRG for the PDB and sequence files respectively;

• either of the two following items to define the missing-atoms distribution:
  • a PDB file specifying the envelope for the whole molecule (fragment + missing atoms), for which we will use the symbol PDBNUP;
  • a CCP4 map file with an electron density in the whole cell, that is going to be used to define the envelope for the whole molecule (fragment + missing atoms), for which we will use the symbol MAPNUP;
  The program can generate its own electron density based on the initial phases and therefore neither PDBNUP nor MAPNUP are strictly speaking necessary in input.

• the chemical composition of the pool of missing atoms;

• the mean electron density. temperature factor, and fractional volume for the solvent region.

Refinement and completion with BUSTER

• Task 0 : Just phasing

  The phases from the partial structure, missing atoms model, bulk solvent and if available, experimental phases, are combined together after a round of scaling to produce BUSTER model Fourier amplitudes and phases in the file mlphas.mtz;

• Task 1 : Maximum-Likelihood partial structure refinement.

  The atomic parameters to be refined are those in PDBFRG. At each cycle BUSTER calculates the log-likelihood L = log L and its derivatives to first and second order w.r.t. the expectation of the total structure factor from the current model. The first-order derivatives (gradients) are the ML equivalents of the weighted difference coefficients used at the corresponding stage of the LS method. Just like the latter they are used by TNT, through system calls from BUSTER, to calculate the gradient of L in atomic parameter space by the Agarwal-Lifchitz method. The second-order derivatives are then used to calculate the (diagonal) curvatures of L in atomic parameter space. These gradients and curvatures are based on the available X-ray data, i.e. experimental amplitudes and optionally phase information. TNT builds similar gradients and curvatures for a geometric restraint criterion and, if required to do so, for other criteria such as compliance with non-crystallographic symmetry. Gradients and curvatures from all sources are then pooled with appropriate weights to build a shift direction in atomic parameter space. The final stage is to determine by a line search how far to move along that direction, then to apply the corresponding shifts and return control to BUSTER to initiate the next cycle. This iterative calculation keeps copious archives of its progress.

• Task 2 : Maximum-Entropy partial structure completion.

  Assuming that the partial structure has already been refined, or that errors due to its imperfection are minor in comparison with those due to its incompleteness, we now request that BUSTER starts modulating the probability distribution of the missing atoms beyond the rather flat "prior prejudice" computed initially as prior by blurring the binary mask for the set-theoretic difference between the whole molecule and the fragment. This is done by selecting a subset of reflexions called the "trunk" (i.e. the unbranched part of a tree) through a combination of resolution and figure-of-merit criteria, then assigning to each trunk reflexion a Lagrange multiplier through which to carry out ME-modulation of the prior prejudice. This iterative process is conducted until convergence is reached, at which stage the initial prior is renamed oldprior and the ME map is written out as the new prior. The evolution of entropy loss and Bayesian score is monitored throughout the calculation.