Browsing the BUSTER output

Maximum Likelihood Refinement

Maximum Entropy Completion

Final Results

• Average Figure of Merit

  This plot shows the average Figure of Merit vs. resolution. It is only output at beginning and at the end of the calculation. Remember that high figures of merit mean a phase distribution that is very sharp, i.e. precise, but not necessarily accurate. In particular, the F.o.M. depends crucially on the variance estimates, and can be overestimated if the variances are too small.

• Structure Factor Amplitudes

  The file contains the plot of average structure factor amplitudes on absolute scale vs. resolution. It can be useful in the diagnostic of scaling problems, e.g. at low resolution where bulk solvent models are often inadequate. It is only output at beginning and at the end of the job. For the definition of the quantities plotted, see below

  Five quantities are plotted vs. the resolution d*:

  1. F_obs: the observed structure factor amplitudes brought to absolute scale by the current values of the scale factor and overall B factor;

  2. F_frg: the structure factor amplitudes for the partial structure in the current fragment file;

  3. F_calc: the amplitude of the total structure factor as calculated from the current model; it has contributions from the partial structure, as well as from the missing random atoms and from the bulk solvent;

  4. F_xpct: the expectation value for the total structure factor amplitude as calculated from the Rice distribution it follows; it depends on the variance as well as on the offset:
     F^xpct_h= ò ^¥ ₀  |F|_h  ×   Rice(|F|_h)  d|F|_h

  5. sigmaF_obs: the estimated standard deviations for the observed structure factor amplitudes; they can be useful to check the take off of the noise with resolution.

The $BDG_job.final.mtz file

The columns in the file are:

	FCTR, PHICTR		Centroid Electron Density Map	:	mFobsexp(i × j centroid)
	2FOFCWT, PH2FOFCWT		SigmaA-weighted 2Fo-Fc	:	2mFobsexp(i × jcentroid)-Fcalcexp(i × jcalc)
	FOFCWT, PHFOFCWT		Fo-Fc difference coefficients	:	mFobsexp(i × jcentroid)-Fcalcexp(i × jcalc)
	FOFRGSLV,PHFOFRGSLV		A second type of Fo-Fc difference coefficients, in which the Fc part has no contribution from the missing atoms model	:	mFobsexp(i × jcentroid)-(Ffragexp(i × jfrag)+Fsolvexp(i × jsolv))
	FOM		BUSTER figure of merit
	HLA HLB HLC HLD		Hendrickson-Lattmann coefficients encoding the BUSTER model phases

In addition to these the file contains the two columns with the experimental amplitudes and sigmas used for the refinement - their labels unchanged.

Notice that the same Fourier coefficients are written to the file shell.01/mlphas.mtz at each refinement cycle (the latter file is overwritten at each refinement cycle).

If Maximum Entropy completion is run after a refinement, the amplitudes and phases at the end of the refinement and before the MaxEnt completion are saved in the MTZ file shell.01/mlphas_beforeME.mtz

Other files containing information which you may wish to consult in order to check the input and the data reside in the "root" directory of the BUSTER tree, the one named <ProjectID>.<run number>. These files are HTML documents, usually accessible from the main output file LIST.html via hyperlinks.

Files in the shell directory

• hist_frg_omega.html and hist_whole_omega.html

  These files contain the cumulative distribution of the omega function for the partial structure and the whole macromolecule, respectively.

  The omega function is either the local fluctuation of the partial structure density or its local average, depending on the parameters chosen in input for the envelope tracing.

  Together with the partial structure volume fraction and the protein volume fraction, these cumulative distributions are used to construct a Fermi-Dirac envelope for the missing atoms: first the Fermi-Dirac envelopes are computed for the partial structure (m_frag(x)) and whole structure (m_whole(x)). Both of these envelopes m(x) are normalised so that 0< m(x) < 1.

  Then, the envelope for the missing atoms is obtained by:

  m_miss(x)=m_whole(x) × [1-m_frag(x)]

• hist_prior.html

  The file contains the cumulative distribution of the missing atoms envelope m_miss ; together with the specified solvent fraction (see the solvent description fields in the input form) the cumulative distribution is used to assess the contouring level at which the prior distribution itself can be displayed as a mask with the desired solvent fraction.

• prior.html

  This file contains information about the generation of the prior envelope. First, a binary mask is traced around the model in the PDB model for the whole molecule, as specified in the PDBNUP file. A 4 Å radius is used as default; the value of this radius can be altered in input by means of the MSKRAD keyword.

  Then, a different procedure is followed, depending on the presence/absence of a know "fragment" substructure:

  • if no fragment is present, the binary mask around the whole molecule is blurred by theta filtering, as described above for the (complement of ) the bulk solvent distribution. The resulting blurred distribution is used as a prior;
  • if a fragment is present, a binary mask is built around the PDB model for the fragment; a logical "whole but not fragment" binary mask operation is performed; and the resulting prior binary mask is blurred, again as described above for the whole

• LogLik0.html

  The file contains the average values of the Likelihood of the starting model (null hypothesis). The total likelihood has been divided by the number of reflexions, so the value should be independent of the number of reflexions.

  Typical values for reasonably good models/data range between -10 and 0. Values smaller than -10 might indicate severely incomplete or imperfect models, or errors in scaling and/or in the internal error model.

• HenLat_<shell#>.html

  The file contains the output from the analysis of the modes of the Hendrickson-Lattmann structure-factor phase probability distribution. The number of zero-, uni- and bi-modal reflexions are listed. The phase probability distribution for a subset of the reflexions whose maxima were rejected is output as a plot of P(j) between 0 and 360 degrees.

• trnk.<shell#>.html

  The file contains the list of reflexions which are accepted into the "trunk", i.e. are assigned Lagrange multipliers in the MaxEnt modulation of the prior prejudice. The figure of merit for these reflexions is higher than the threshold set with the FOMTHR field/keyword. The resolution limits are those set via the Maximum Entropy resolution limit field at input time.

  In the course of the maximum Bayesian score modulation of the prior prejudice, one or two Lagrange multipliers will be varied respectively for each centric or acentric reflexion in the list.

• unimod.<shell#>.html

  The file contains a summary of the Bayesian score maximisation that leads to the Maximum Entropy map for the missing structure: Lagrange multipliers for reflexions in the trunk are varied, and structure built into the initial envelope (prior).

  The convergence criterion is echoed first. Iteration is terminated when the shift in Bayesian score from one cycle to the next (normalised to the value of the Bayesian score itself: see second column in this file) falls below this value. The threshold has a default value of 10E-05.

  Then, at each cycle of the iterative maximisation process, the following quantites are reported:

  • Bayesian score fractional change: it is the ratio between the total variation in Bayesian score from the first to the current cycle, and the current value of the Bayesian score itself;
  • Bayesian score fractional shift: it is the ratio between the variation in Bayesian score from the previous to the current cycle, and the current value of the Bayesian score itself. It is the quantity used as a termination criterion for the iteration;
  • BS: Bayesian Score, defined as:
    BS=LogLik + N S_m

    where N is the number of random scatterers, as declared via the Atomic Composition field in the input form;

  • S_m: Entropy Loss, referred to the starting prior m(x);
  • Grad S: components of the gradient of the Entropy Loss with respect to the two Lagrange multipliers associated to the distance restraint and to the Entropy loss;
  • Grad L: components of the gradient of the Log-Likelihood with respect to the two Lagrange multipliers associated to the distance restraint and to the Entropy loss;
  • Hess S: elements of the Hessian of the Entropy Loss with respect to the two Lagrange multipliers associated to the distance restraint and to the Entropy loss;
  • Hess L: elements of the Hessian of the Log-Likelihood with respect to the two Lagrange multipliers associated to the distance restraint and to the Entropy loss.

• qadsol.<shell#>.html

  The file contains a detailed summary of the Bayesian score maximisation that leads to the Maximum Entropy map for the missing structure.

  The Likelihood-Gain is at each step maximised with a constraint of the Loss of Entropy and a constraint on the shift ('Distance') in Bayesian Score: two Lagrange multipliers Mu and Nu are associated respectively with the Entropy and distance constraints.

  The actual number of dimensions of the subspace in which each cycle takes place is output (SD) together with the total number of dimensions of the space (TD, always equal to 2).

  The shifts along the two directions corresponding to the Entropy Loss and the Log-Likelihood distance Lagrange multipliers are also reported under the headings DirS Shift and DirL Shift.

• BS_surf.<shell#>.html

  The file contains information about the trajectory along the Bayesian Score surface, as it is waded through during the modulation of the prior-prejudice. The maximisation is really carried out as a minimisation of the negative of the Bayesian Score, hence the minus signs in front of the BS columns headings.

  At each cycle an attainable value of the Bayesian Score is echoed ( -BS Att.), together with the distance from the current point to the attainable BS (D Attainable); the distance constraint will sometimes allow only a closer value of BS to be targeted ( -BS All.); the distance to this target value is reported in the last column ( D to -BS Target).

• mlnorm_unmd.<cycle#>.html

  At each cycle during the modulation of the prior-prejudice, the Log-Lik Gain, the overall scale and B factors and the fragment imperfection B factor [2,3] are reported. For details, see below the documentation given for the files mlnorm.<cycle#>.html.

• HLplots directory

  The shell.01/HLplots folder contains the output files from the analysis of the Hendrickson-Lattmann phase probability distribution; the analysis is done to gather unimodal reflexions in the "trunk" used to calculate the Maximum Entropy map during completion jobs. BUSTER rejects maxima of the phase probability for some reflexions, either on the grounds that they are too shallow, or merged into a single peak, or but a minor hump with respect to the main maximum. The phase probability distribution for a subset of the reflexions whose maxima were rejected is output as a plot of P(j) between 0 and 360 degrees.

• rmsdrho.mtv

  This file is output when the input MTZ file contains FOM's. It displays the rmsd of the electron density computed from the current centroid phases as from (Blow and Crick, 1959):

  <Dr**2> = S _h[ e_h ×  (1-FOM_h**2)  ×  F_h^obs**2 ]

• Fragments directory

  The shell.01/Fragments folder contains the TNT coordinate files ( shell.01/Fragments/cycle_<cycle#>.cor) at the various cycles during the course of the refinement. The coordinate files can be deleted when archiving the output with the Save option, depending of what you specify at the end of the preferences.

  If you wish to transform the TNT coordinate file into a PDB file, you can either do it via the map display page, or run $tntbin/convert:

  			$tntbin/convert << eof
  			CELL a b c alpha beta gamma
  			INCLUDE <$tntdata>/connect.dat
  			INCLUDE cycle_00n.cor
  			INCLUDE sequence.file.seq
  			PUNCH cycle_00n.pdb BROOKHAVEN
  			eof
  			

  The partial structure model at the last refinement cycle is output to the PDB file <JobName>/shell.01/<JobName>.final.pdb.

TNT output files

Three TNT output files contain the output from the script tntlongll:

1. agarwal.dat

   Output file from the call to the TNT module rfactor that computes the gradients for the working set reflexions.

2. rfactor.old

   Output file from the call to the TNT module rfactor that computes the curvatures for the working set reflexions.

3. rfactor.dat

   Output file from the call to the TNT module rfactor that merges the gradients and curvatures for the working set reflexions.

geometry.dat is the output file from the call to the TNT module geometry that computes the gradients and curvatures of the GEOM stereochemical restraint criterion.

geometry.html is the output from the geometry module of TNT, and lists the worst violations to the geometric restraints.

Three ASCII hkl files in TNT format are stored in the <ProjectID>.<run number> directory of the BUSTER job directory tree:

1. tnt.hkl The file contains the structure factor amplitudes in TNT format, as read from the input MTZ file, and after the rejection of the data listed in the file rejections.html.

2. tnt_<FreeR_flag>.hkl The file contains the amplitudes with the FreeR_flag chosen in input, which is used as a test set for cross validation. The default test set is the one with FreeR_flag=0.

3. tnt_no<FreeR_flag>.hkl The file contains all the data but the test set FreeR_flag, and is therefore the working set for structural refinement.

View Maps

Types of maps
	BUSTER computes Fourier coefficients for five kinds of maps: Centroid Electron Density Map The centroid electron density map is obtained by Fourier Transform of observed structure factor amplitudes, and the centroid phases (as introduced by (Blow and Crick, 1959)) derived from the phase probability distribution corresponding to the current statistical model. Therefore, the centroid map contains the electron density for the atoms in the fragment, and for the current model for the random atoms and solvent. It is less noisy than the sigmaA-weighted 2Fo-Fc (see below) but not as sensitive to details in the missing part of the structure. "SigmaA" weighted 2Fo-Fc Electron Density Map The map is a 2Fo-Fc electron density (see Main and Read); but the BUSTER map uses the current phase probability distribution which does possess non-Wilsonian contributions from the missing atoms). The map is produced by Fourier transforming the coefficients 2FOFWT and 2PHFOFCWT in the $BDG_job.final.mtz file in the shell.01 directory. Keep in mind that the centroid should be less noisy, but that fragment and missing features are put on approximatively the same scale in the "sigmaA" map, so the latter is the map to inspect for weakly scattering missing structure. "SigmaA" weighted Fo-Fc Difference Density Map The map is produced by Fourier transforming the coefficients FOFCWT and PHFOFCWT in the $BDG_job.final.mtz file in the shell.01 directory. Here, Fc=Ffrag+Fmiss+Fsolv The positive (negative) contours of this map will show regions where the current model lacks (has too much) density, and the negative regions will be regions where the model of partial structure, missing atoms and solvent has too much (little) density, either because: a partial structure atom is misplaced or its B factor being too small (high); or the prior prejudice distribution for the missing atoms is too strong (weak); or the solvent envelope is misplaced. This map is identical to the Fo-Ffrag-Fsolv described below if there are no missing atoms declared. "SigmaA" weighted Fo-Ffrag-Fsolv Difference Density Map The map is produced by Fourier transforming the coefficients FOFRGSLV and PHFOFRGSLV in the $BDG_job.final.mtz file in the shell.01 directory. Notice that the model Fmiss for the missing structure is not subtracted from the Fo here. The positive contours of this map will show regions where the current model lacks density, either because a partial structure atom is misplaced or its B factor being too small; or because there are missing atoms; or because the solvent envelope is misplaced. This map is identical to the Fo-Fc described above if there are no missing atoms declared. Maximum Entropy Map for the missing part of the structure The file you obtain at the end of a structure completion job (Maximum Entropy structure completion button) is stored in the shell directory of your current run, under the name Q_Map.<cycle#>.map (e.g. busterfiles/logfiles/PROTEIN.4/shell.01/Q_Map.022.map). Notice that this map is a positional probability distribution for the missing structure only, and does therefore look peaky and sometimes not well connected... a more 'traditional' map for the missing part can be inspected looking at the positive contours of the Fo-Fc map. You can display the Maximum Entropy map by clicking on the View Maps hyperlink and select the Maximum Entropy map as the object to be displayed. This will trigger scripts that will service it to the displaying tool of your choice. You can extend your fragment by building or rebuilding into that MaxEnt map at least some of the hitherto missing structure. The Fourier coefficients for the Centroid and sigmaA weighted maps are all in the file shell.01/$BDG_job.final.mtz. In addition, the Maximum Entropy map is produced in CCP4 format, and is contained in the shell.01 directory, in the file Q_Map.<cycle#>.map
Displaying maps
	From the Plotting page, you can select the maps and model to be displayed. The displaying can be done in two ways: Plot: this option will trigger the generation of maps and models on the server machine; a gzipped tar file is then serviced to the client machine, and the application specified in your .mailcap file will display the results on the client; Script: this option will generate a script that you can save to disk, and execute from the client machine; the script will generate the maps and model and display them on the client; this is especially useful when the maps are so big that the conventional plotting route would lead to time-out of the plotting request on the part of the browser. The Save Map(s) and PDB file(s) (tar file) option allows you to save to disk a tar file with your map(s) and model(s), in CCP4 and PDB format respectively. Note :  for the display tools to work, the paths to the helpers binaries NEED to be set by the system administrator in the file: $BDG_home/bin/helpers.local for all the client machines you want to display results on. Furthermore, your browser needs to have the correct handling enabled for these tools. Please see the installation instructions for the Helper applications. You have the choice between different viewpoints: examining the density around the PDB models, the whole cell or just the asymmetric unit. The default can be set editing the preferences from the BUSTER Control panel. Each viewpoint has its advantages: The "asymmetric unit" plot contains all the information, takes less time to compute and uses less space in memory and on disk. If your viewer supports this, crystallographic symmetry information can be used to generate density at any point in space using this unique volume of density. "Contour around the model": any PDB file in your datafiles directory whose name starts with "model_" and ends with a ".pdb" suffix will be eligible for display at this time, e.g. "model_protein.pdb". The contouring around the model will allow for inspection of the connectivity of the map when the model sits across the asymmetric unit boundaries. If more than one model is present, the last one will be selected for the contouring. The "whole cell" option enables you to see the molecular boundaries and the chain connectivity without bumping into the limits of the asymmetric unit.

References

1. Blow, D.M. and Crick, F.H.C. (1959). Acta Cryst. 12, 794-802.
2. Bricogne, G. (1993). Int. Tab. for Cryst. B, 23-106.
   
3. Bricogne, G. & Irwin, J.J. (1996) in Macromolecular Refinement - Proceedings of the CCP4 Study Weekend. SERC Daresbury Laboratory, Warrington, England; 85-92.
4. Bricogne, G. & Irwin J.J. (1997). In Crystallographic Computing 7 edited by P.E. Bourne and K.D.Watenpaugh, 1-9.
   
5. Diamond, B. (1993). Int. Tab. for Cryst. B, 345-373.
   
6. Main, P. (1979). A Theoretical Comparison of the Beta, Gamma' and 2F_o-F_c syntheses.Acta Cryst., A35, 779-785.
   
7. Read, R. (1986). Improved Fourier Coefficients for Maps Using Phases from Partial Structures with Errors. Acta Cryst., A42, 140-149.
   
8. Collaborative Computational Project, Number 4 (1994). Acta Cryst. D50, 760-763.