research papers
Analysis and validation of macromolecular B values
^{a}Institute of Molecular Biology and Biotechnology ANAS, Matbuat Avenue 2a, Baku 1073, Azerbaijan, and ^{b}MRC Laboratory of Molecular Biology, Francis Crick Avenue, Cambridge CB2 0QH, England
^{*}Correspondence email: garib@mrclmb.cam.ac.uk
This paper describes a global analysis of macromolecular B values. It is shown that the distribution of B values generally follows the shifted inversegamma distribution (SIGD). The parameters of the SIGD are estimated using the Fisher scoring technique with the expected Fisher information matrix. It is demonstrated that a contour plot based on the parameters of the SIGD can play a role in the validation of macromolecular structures. The dependence of the peakheight distribution on resolution and atomic B values is also analysed. It is demonstrated that the Bvalue distribution can have a dramatically different effect on peak heights at different resolutions. Consequently, a comparative analysis of the B values of neighbouring atoms must account for resolution. A combination of the SIGD, peakheight distribution and outlier detection was used to identify a number of entries from the PDB that require attention. It is also shown that the presence of a multimodal Bvalue distribution often indicates that some loops or parts of the molecule have either been mismodelled or have dramatically different mobility, depending on their environment within the crystal. These distributions can also indicate the level of sharpening/blurring used before atomic structure It is recommended that procedures such as sharpening/blurring should be avoided during although they can play important roles in map visualization and model building.
Keywords: macromolecular structures; B values; inversegamma distribution; maximumlikelihood estimation.
1. Introduction
) are essential steps in the derivation of reliable threedimensional structures of macromolecules. Atomic procedures based on Bayesian statistics are now routine (Bricogne, 1997; Murshudov et al., 2011; Pannu & Read, 1996). Prior structural and chemical information pertaining to building blocks of macromolecules are used during (Vagin et al., 2004; Long et al., 2017; Moriarty et al., 2009; Nicholls et al., 2012; Smart et al., 2011, 2012) as well as for validation (Davis et al., 2007; Read et al., 2011). This aids the derivation of chemically and structurally sensible atomic models that are consistent with prior knowledge, whilst transferring as much information from the experimental data to the model as possible via the likelihood function. There are a number of research papers and software tools that are dedicated to the validation of positional parameters of atomic models (see Chen et al., 2010; Hooft et al., 1996; Read et al., 2011, and references therein). These papers and the corresponding software tools have been instrumental in improving the quality of the models deposited in the PDB (Read et al., 2011; Berman et al., 2002). As highlighted by Pozharski et al. (2013) and Weichenberger et al. (2013), there is still a long way to go before we can claim that the quality of the models in the PDB agrees well with prior knowledge and optimally reflects experimental data.
and validation of atomic models elucidated using crystallographic, and more and more increasingly singleparticle cryoEM, methods (Frank, 2006Parameters of atomic models include positions, occupancies and isotropic/anisotropic atomic displacement parameters (ADPs; also known as B values, U values and temperature factors). While many research papers have been dedicated to the validation of ADPs (see Read et al., 2011, and references therein), works dedicated to the analysis of ADPs are few and far between (Dauter et al., 2006; Merritt, 2011, 2012; Negroni, 2012;). Moreover, ADPs are often used as a proxy for model quality, for example for the selection of reliable portions of macromolecules, and for validation (Chen et al., 2010). There are at least two problems with such procedures.

The resolution at which the structure has been resolved should also play a role in the analysis of B values. For example, a Bvalue difference of 10 Å^{2} would have a completely different effect on the shape and height of the density were the data to extend to a resolution of 3 Å compared with 2 Å. At 3 Å resolution this 10 Å^{2} difference would have little effect on the shape and height of the density, whereas at higher resolutions a difference of 10 Å^{2} would have a more dramatic effect. The relative difference between B values is also of importance; the effect of atomic Bvalue differences of 10 versus 20 Å^{2} will be more dramatic than 100 versus 110 Å^{2}. Moreover, the distribution of B values should be considered in the analysis of B values and the selection of reliable portions of the structure. One should always remember that oversharpening can make some ADPs negative, making them physically unreasonable. On the other hand, too much blurring may result in loss of structural details in the density maps.
In this contribution, we will describe some statistical tools for the analysis of isotropic B values. Firstly, we explore using the shifted inversegamma distribution (SIGD) to describe the distribution of ADPs within one as well as for each chain within the molecule. We then analyse the distribution of atomic peak heights for point atoms at different B values and resolutions, and then describe the application of these techniques to selected entries from the Protein Data Bank (PDB; Berman et al., 2002).
It should be noted that the inversegamma distribution has previously been suggested (Dauter et al., 2006) and used (Negroni, 2012) for the analysis of macromolecular structures. Here, we extend these analyses and add new statistics: an atomic peakheight distribution for point atoms that is dependent on resolution and B value, which reflects the effect of B values on the density better than mere B values. We also present a new plot based on SIGD parameters that helps in the visual identification of suspect crystal structures.
2. Shifted inverse gamma as a model for the Bvalue distribution
Individual atomic ADPs are proportional to the variances of positional parameters; ADPs represent the mobility of atoms as well as the accuracy of the positional parameters. In Bayesian statistics, the inversegamma distribution (IGD) is often used as a prior probability distribution to model the variance of a normal distribution (see, for example, O'Hagan, 1994). Since ADPs are proportional to the variances, we can postulate that it is likely that the distribution of ADPs will resemble the IGD. However, since average B values depend on the sharpening level of the data, we add an additional shift parameter to the IGD. Sharpening/blurring should change the average B value without affecting the shape of the distribution, except in cases where oversharpening produces negative or many small B values. Therefore, we assume that the distribution of isotropic ADPs can be modelled using the SIGD,
This distribution has three parameters: shape (α), scale (β) and shift (B_{0}). If there is no over/undersharpening of Fourier coefficients then B_{0} = 0, although this is rarely the case. Changing ADPs from B to u = B/8π^{2} only affects the scale and shift parameters. Since the shape parameter is also known as the it is tempting to assume that since the positional parameters of atoms reside in a threedimensional space, the shape parameter of the SIGD would be around 3. However, we refine this parameter using soft harmonic restraints to ensure that the estimation of SIGD parameters is stable while allowing some variability.
Fig. 1 shows the SIGD for different shape and scale parameters (Fig. 1a) and the empirical of B values for an example from the PDB (PDB entry 1a4i; Figs. 1b and 1c, corresponding to restrained and unrestrained respectively). This figure demonstrates that, at least for this particular case, the empirical Bvalue distribution resembles the SIGD reasonably well. Comparison of Figs. 1(b) and 1(c) exemplifies how unrestrained does not change the distribution of B values dramatically, at least at this resolution, in this case.
2.1. of inversegamma parameters with harmonic restraints
For each PDB entry in the test set, we refined the SIGD parameters by et al., 1999; Steiner et al., 2003). The derivation of the equations for this case is given in Appendix A. Starting values for the parameters are derived using the relationship between the parameters of the SIGD and its mean and variance. Parameter estimation using the first few moments of the distribution is known as the method of moments (Stuart et al., 1999),
estimation using the Fisher scoring method with the expected Fisher information matrix (StuartNote that the mean of the SIGD exists only if α > 1, and the variance exists only when α > 2. We have two equations and three parameters to estimate. We choose the starting value of B_{0} to be equal to 90% of the minimum of the ADPs in the PDB file considered, thus reducing the number of unknown variables to two. Equations (2) are solved to find the initial values as
These starting values are iteratively improved using Fisher's scoring method (Appendix A) until convergence is reached. During the estimation process, care was taken to make sure that the estimates are reasonable, i.e. shape parameters are restrained to be near 3.5 and shift parameters are constrained to obey B_{0} < 0.9B_{min}. Note that a negative value of B_{0} is an indication that the observed data may have been oversharpened beyond reason, in which case parameter estimation becomes unstable.
3. Resolution and Bvaluedependent peak height at the centre of atoms
Since the 3D maps used for model building in crystallographic and cryoEM experiments correspond to densities of electrons and electrostatic potentials, respectively, it is interesting to analyse the effect of B values on the peak height at the centre of atoms for a given resolution. It is clear that the peak heights are dependent on atom types, occupancies, resolution and B values. In order to allow comparison of atomic peak heights, we ignore the effect of different atom types and occupancies; we essentially treat all atoms as point atoms^{1}. As a result of resolution cutoff and B values (atomic mobility), the density becomes smeared out; this affects the values of the density map at the centre of atoms. The density corresponding to the point atom with B value equal to B_{mod} and resolution s_{max} = 1/d_{max} can be calculated using (see, for example, Chapman, 1995)
This is the shape of the pointatom density corresponding to a given resolution and B value. The density at the centre can be calculated by letting x = 0:
As can be seen, the density at the centre of the atoms depends on the resolution as well as on the B value. The real observed density will also depend on the noise level, the weights used in map calculations, the occupancies of the atoms, the quality of the amplitudes and phases, the number, types and proximity of neighbouring atoms, the overall anisotropy of the data and many other factors. However, a very simple analysis of peak heights should shed some light onto what can be expected at a given resolution. Even if the distribution of the B values is known, it is tricky to derive a closedform expression for the distribution of peak heights at the atomic centre; therefore, in the following analysis we will use only empirical and simulated distribution histograms for peakheight analysis.
Fig. 2 shows the histogram of B values for one example and the corresponding peakheight distribution calculated at different resolutions. It is seen that although the Bvalue distributions are the same, changing the resolution dramatically changes the distribution of peak heights. At higher resolution, as might be expected, small differences in B values have a much more dramatic effect on the density level than at lower resolutions. Therefore, it is always a good idea to consider the Bvalue distribution together with resolution, or to consider how the Bvalue distribution and its effect on peak heights depend on resolution.
4. Selection of the PDB entries
We considered the 89 862 entries from the PDB, as of December 2016, for which the experimental method was Xray crystallography. For further analysis, we used only the models for which the highresolution diffraction limit is between 1.5 and 3 Å. To avoid dealing with REFMAC5 (Kovalevskiy et al., 2018) available from CCP4 (Winn et al., 2011). Reasons for failure include (i) the ligand present in the PDB was not in the CCP4 monomer library at the time of (this was the most common case), (ii) no structurefactor amplitudes^{2} and (iii) spacegroup inconsistencies between the PDB file and the reflectiondata file. The remaining crystal structures contained roughly 160 000 chains, among which there were 145 800 protein chains. We also excluded cases with R factors higher than 35%. We used these crystal structures and corresponding chains for further analysis. Table 1 gives a short summary of the selection of PDB entries.
constraints, the use of which is not always clear from the PDB entry, we removed virus structures. Of the remaining models, we were able to refine 46 952 automatically using the program

Table 2 lists the entries used as examples in this work. It should be stressed that the aim of this contribution is not to criticize a particular PDB entry; rather, we would like to highlight the shortcomings of the techniques used at the time of elucidation of these macromolecular structures.

5. Results and discussions
5.1. Overall parameters of the shifted inversegamma distribution
We applied the method described above for the estimation of SIGD parameters for all selected PDB entries. The method was implemented in the statistical package R (R Core Team, 2018) and was later ported to Python. Figs. 3(a) and 3(b) show the shape and scale parameters versus resolution. For completeness, we also show average B value versus resolution (Fig. 3c). It is evident that both α and β depend on resolution.
The plot of α versus β^{1/2} (Figs. 3d and 3e) shows the codependency of these two parameters. Moreover, very high values of α or β are an indication that either suboptimal was performed or the data have been massaged (whether intentionally or not) before was performed. If α and β are outside the region defined in Fig. 3(e) then the model should be considered for rerefinement with careful consideration of atom types as well as potentially erroneous loops. We find this plot to provide useful information regarding behaviour. In all our following analyses we use this plot to identify peculiar PDB entries.
5.2. Oversharpened cases
When the Bvalue distribution is shifted towards the left with a floor at a certain B = B_{min} and a large number of atoms all have the same small B value (Fig. 4d) then it is an indication that the data have been oversharpened before In this case, the parameter α becomes small or of the SIGD parameters becomes unstable. There are a few such cases in the PDB; here, we use PDB entry 3ad4 as an example. Fig. 4 shows the SIGD parameters on the α/β plot (Fig. 4a) and the histogram with the associated SIGD distribution (Fig. 4b), as well as the corresponding peakheight distribution (Fig. 4c) for this case.
For the purposes of visualization and model building, slight oversharpening can be an acceptable procedure if care is exercised with the interpretation of water molecules; the effects of series termination and noise amplification can result in peaks that could be incorrectly interpreted as waters. In such cases, a careful analysis of contacts must be performed in order to avoid modelling wrong `water' molecules. However, for B values. Almost all existing programs use some lower threshold (for example 0.5–2 Å^{2}) in order to avoid dealing with negative B values. As a result, the Bvalue distribution becomes distorted, making interpretation of the B values (and sometimes the whole molecule) difficult. When TLS is also performed then the behaviour of B values can become even more unpredictable, resulting in large negative or positive peaks reflecting wrong atomic B values.
oversharpening can result in suboptimal atomic models as some of the atoms may have negativeFor our example case, we added a small B value (20 Å^{2}) and rerefined the model, resulting in the Bvalue distributions shown in Figs. 4(d), 4(e) and 4(f), with α and β parameters falling in the region identified in the α/β plot.
5.3. Blurred cases
Blurring, in general, should not result in suboptimal B_{0}.
However, blurring would result in smeared and convoluted maps that would be difficult to interpret. Moreover, FFTbased programs would become slow and derivative calculations would be inaccurate. If is carried out carefully (it is software dependent) then the only parameter that would be affected isIn general, the minimal B value is the safest level that can be used for map sharpening and/or However, it is not easy to derive the minimal B value before atomic Such sharpened maps (using minimal B as a sharpening parameter) would show more detail without causing negative B values.
An example of an overblurred case is PDB entry 5g4t (Fig. 5). In such cases, subtracting a B value slightly less than the minimum B value for can change behaviour.
5.4. Outliers
There are a number of cases with large α or β values. These are difficult to classify, and there may be various reasons for these outliers. One such case is PDB entry 2bp7: from viewing the parameters of the estimated SIGD on the α/β plot (Fig. 6a), it is evident that this is a clear outlier with a larger α of 7.83 and a relatively large β of 308.21. The SIGD (Fig. 6b) and peakheight distributions (Fig. 6c) show that there are some atoms with small B values and therefore relatively large peak heights. Closer analysis revealed that most of the side chains of methionine residues are missing, and that there are some heavier atoms that are interpreted as water molecules. The crystallization conditions indicate that there could be magnesium, sodium, potassium, chlorine and phosphate present as well as molecules of thiamine diphosphate. Inspection of the map showed that these molecules/ions are present in the crystal. After a few rounds of model building and rebuilding, R and R_{free} decreased from 0.197 and 0.245 to 0.160 and 0.225, respectively (see supporting information). The SIGD parameters estimated using B values from the rebuilt structure are close to the `acceptable' region of the α/β plot (Fig. 6d). The histogram of the B values, together with the fitted SIGD (Fig. 6e) and peakheight distribution (Fig. 6f), show that there are still some atoms with small B values as well as some with large B values that need to be removed. Figs. 6(g) and 6(h) show the region of the density before and after rebuilding in the region of the thiamine diphosphate. This example demonstrates how the Bvalue distribution, together with the fitted SIGD parameters, can help to identify some of the shortcomings of atomic models. In particular, very low B values can indicate that at least some of the atomic identities are wrong.
5.5. Multimodal Bvalue distributions
There are a number of PDB entries for which the distribution of B values exhibits multimodality. There can be various reasons for such behaviour. Two such reasons are the following.

Here, we analyse two cases: PDB entry 2grm, where there are two sets of subunits with very different B values (Fig. 7), and PDB entry 4l39, where the density for part of the molecule is very weak (Fig. 8).
For PDB entry 2grm (Fig. 7), the Bvalue and peakheight distributions clearly exhibit multimodal distributions (Figs. 7b and 7c). The parameters of the SIGD for this case are slightly outside the α/β plot (Fig. 7a). Analysis of the B values shows that chains A and B have low B values and chain C has higher B values. The density corresponding to chains A and B is also better than that for chain C (Figs. 7d and 7e). In this case, it seems that the occurrence of different B values for different chains is owing to the fact that they have a different number of contacts to stabilize them within the crystal (Table 3). Chain C interacts only with itself within the and makes only 15 interactions with chain B. In contrast, chains A and B interact with each other within and between asymmetric units. Presumably, this is the reason why the mobility of chain C is higher than those of the others, resulting in higher overall B values. The peakheight distribution (Fig. 7c) also shows that there is at least one atom that is substantially heavier than the other atoms. Analysis of the density revealed that this atom is on a twofold axis. The PDB header shows that lithium chloride was used for crystallization. If this atom is heavier than the other atoms then it is likely to be a Cl atom, although the identification of Cl atoms is usually very difficult, especially at this resolution (2.8 Å).

In the case of PDB entry 4l39 (Fig. 8), the higher B values, and therefore the more smeared density, are the result of a disordered domain. Inspection of the density (Fig. 8d) clearly indicates that there might be a different interpretation of this domain and that it might perhaps be better if such loops were left out of the atomic model altogether.
6. Conclusions and future perspectives
We have demonstrated that there is a need to model as well as to validate atomic ADPs. It is demonstrated that for many macromolecular structures the SIGD can be used to model the distribution of ADPs. Even if the Bvalue distribution over the whole structure does not obey the SIGD, the individual chains/domains will obey this distribution. When the distributions of B values for different chains/domains are different there can be at least two reasons: (i) different domains/subunits have different contacts depending on the environment and (ii) there are disordered and/or mismodelled regions that have naturally higher B values, reflecting errors in the positional parameters. Such multimodality affects the density and therefore the interpretability of the maps. Future work will include the of multimodality parameters (the number of classes and parameters of the SIGD for each class) using such techniques as the expectationmaximization algorithm.
We also show that by modelling the Bvalue distribution using the SIGD and comparing the parameters with those derived from all PDB entries, one can identify the degree of sharpening/blurring before Whilst both sharpening and blurring are valid procedures that can help in the interpretation of density maps, applying these procedures prior to or during can result in suboptimal atomic models. Too much sharpening results in seriestermination and noiseamplification effects that reduce the interpretability of maps, and there might be additional density owing to seriestermination effects that can be erroneously modelled as water molecules. In general, it is recommended that during map interpretation, even if sharpening is used, the local contacts must be analysed in order to ensure that the modelled atoms/molecules make chemical sense. During a reasonable level of sharpening must be used, otherwise can become very slow and convergence may not be reached, or can result in negative B values that consequently may affect the density map as well as the procedure.
Since the signaltonoise ratio in the maps is related to the overall mobility, and therefore the atomic B values of the molecules, it is not surprising that when the distribution of the B values for a given subunit/domain is shifted to the right (increased) then the corresponding density becomes weaker and the resolvability of peaks is reduced. Estimation of the SIGD parameters can also provide an estimate of the minimal safe B value for sharpening. When the data are oversharpened this affects the distribution of B values, moving them to the left (reducing them). In this case, the procedure often becomes stuck, as B values cannot become negative. Correcting for oversharpening, i.e. artificially adding a B value to the data before seems to improve the behaviour of leading to a more reliable atomic model with associated ADPs.
In this work, we also analyse the resolutiondependent peakheight distribution and show that the effect of the Bvalue distribution is dramatically different at different resolutions. In principle, the Bvalue distribution, together with its effect on the peakheight distribution, can be used as a proxy for the resolution/information content of the data: at higher resolution it can be expected that the Bvalue distribution will be narrow and at low resolution it should be wider.
In future, we plan to apply a theoretical Bvalue distribution for the of atomic models at lower resolutions where the data are not sufficient to accurately estimate individual atomic B values.
Future work will also consider the potential of using the IGD for map sharpening. The usual mapsharpening technique assumes that there is a single B value for all atoms. If we use the distribution of B values then we should be able to design better sharpening normalization procedures and therefore better mapsharpening procedures. A similar approach was used by Blessing et al. (1996), under the assumption that the Bvalue distribution follows a normal distribution. We also intend to include local Bvalue analysis together with the effect on peak heights to identify wrongly modelled atoms or regions of the structure.
APPENDIX A
Derivatives and expected Fisher information matrix for estimation of the parameters for the inversegamma distribution
If the distribution of observed data, e.g. B values, is a shifted inversegamma distribution then, for a given sample of size N, the likelihood function will have the form
where the B_{i} are the B values of atoms and α, β and B_{0} are the parameters of the SIGD. Taking the logarithm of this expression and multiplying by −1 results in the negative loglikelihood function,
We want to use Fisher's scoring method with the expected Fisher information matrix to find the minimum of the negative loglikelihood function (Stuart et al., 1999; Steiner et al., 2003). The first derivatives of this function with respect to the parameters of the distribution have the form
In the above expression ψ(α) = Γ′(α)/Γ(α) is the psi function, which is the derivative of the log of the gamma function. The first derivative of the psi function is called the digamma function, with further derivatives being called polygamma functions (Abramowitz & Stegun, 1964). In principle, these derivatives should be sufficient to use firstorder methods such as the steepestdescent and conjugategradient methods for parameter estimation. However, to use the secondorder methods we need to calculate the second derivatives. We prefer using the Fisher scoring method with the expected Fisher information matrix, the main reason being that the Fisher information matrix is nonnegative. As a result, shifts based on the Fisher information matrix are always directed towards the minima of the negative loglikelihood function.
Let us use the notation I for Fisher's expected information matrix, with elements I_{ij} = E(∂^{2}l/∂p_{i}∂p_{j}) (Stuart et al., 1999) with (p_{1}, p_{2}, p_{3}) denoting the (α, β, B_{0}) parameters of the IG. It can be shown that elements of the information matrix can be calculated using
For the derivation of these formulae, we used expressions for the moments of the IGD: E(x^{n}) = β^{n}Γ(α + n)/Γ(α) valid for n < α, noting that this formula is valid for all negative values of n.
Supporting information
Refined and rebuilt atomic model for 2bp7. DOI: https://doi.org//10.1107/S2059798319004807/qh5059sup1.txt
Footnotes
^{1}For actual one needs to consider the form factors, occupancies, resolution and B values of the atoms. However, in this work, our aim is to analyse the relative effects of B values and resolution; effectively we normalize the peak heights by the atom types and occupancies, i.e. divide the actual peak heights by the occupancies and a factor related to the form factors.
^{2}At the time of the analysis the program REFMAC5 would not use intensities automatically. Since the number of cases in which amplitudes were available was deemed to be sufficient for our analysis, the cases in which only intensities were available were rejected.
Acknowledgements
The authors thank MRC–LMB, Cambridge, England and IMBB of ANAS, Baku, Azerbaijan for creating encouraging research environments. We also thank Rob Nicholls for critical reading of the manuscript and for giving useful suggestions.
Funding information
This work was supported by an MRC grant (MC_US_A025_0104) and Azerbaijan National Academy of Sciences grant decree No. 5/9 dated 15.03.2017.
References
Abramowitz, M. & Stegun, I. (1964). Handbook of Mathematical Functions. New York: Dover. Google Scholar
Berman, H. M., Battistuz, T., Bhat, T. N., Bluhm, W. F., Bourne, P. E., Burkhardt, K., Feng, Z., Gilliland, G. L., Iype, L., Jain, S., Fagan, P., Marvin, J., Padilla, D., Ravichandran, V., Schneider, B., Thanki, N., Weissig, H., Westbrook, J. D. & Zardecki, C. (2002). Acta Cryst. D58, 899–907. Web of Science CrossRef CAS IUCr Journals Google Scholar
Blessing, R. H., Guo, D. Y. & Langs, D. A. (1996). Acta Cryst. D52, 257–266. CrossRef CAS Web of Science IUCr Journals Google Scholar
Bricogne, G. (1997). Methods Enzymol. 276, 361–523. CrossRef CAS PubMed Google Scholar
Chapman, M. S. (1995). Acta Cryst. A51, 69–80. CrossRef CAS Web of Science IUCr Journals Google Scholar
Chen, V. B., Arendall, W. B., Headd, J. J., Keedy, D. A., Immormino, R. M., Kapral, G. J., Murray, L. W., Richardson, J. S. & Richardson, D. C. (2010). Acta Cryst. D66, 12–21. Web of Science CrossRef CAS IUCr Journals Google Scholar
Dauter, Z., Murshudov, G. N. & Wilson, K. S. (2006). International Tables for Crystallography, Vol. F, edited by M. G. Rossmann & E. Arnold, pp. 393–402. Chester: International Union of Crystallography. Google Scholar
Davis, I. W., LeaverFay, A., Chen, V. B., Block, J. N., Kapral, G. J., Wang, X., Murray, L. W., Arendall, W. B., Snoeyink, J., Richardson, J. S. & Richardson, J. S. (2007). Nucleic Acids Res. 35, W375–W383. Web of Science CrossRef PubMed Google Scholar
Emsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. (2010). Acta Cryst. D66, 486–501. Web of Science CrossRef CAS IUCr Journals Google Scholar
Frank, J. (2006). ThreeDimensional Electron Microscopy of Macromolecular Assemblies: Visualization of Biological Molecules in Their Native State. Oxford University Press. Google Scholar
Hooft, R. W. W., Vriend, G., Sander, C. & Abola, E. E. (1996). Nature (London), 381, 272. CrossRef PubMed Web of Science Google Scholar
Kovalevskiy, O., Nicholls, R. A., Long, F., Carlon, A. & Murshudov, G. N. (2018). Acta Cryst. D74, 215–227. Web of Science CrossRef IUCr Journals Google Scholar
Long, F., Nicholls, R. A., Emsley, P., Gražulis, S., Merkys, A., Vaitkus, A. & Murshudov, G. N. (2017). Acta Cryst. D73, 112–122. Web of Science CrossRef IUCr Journals Google Scholar
McNicholas, S., Potterton, E., Wilson, K. S. & Noble, M. E. M. (2011). Acta Cryst. D67, 386–394. Web of Science CrossRef CAS IUCr Journals Google Scholar
Merritt, E. A. (2011). Acta Cryst. A67, 512–516. Web of Science CrossRef IUCr Journals Google Scholar
Merritt, E. A. (2012). Acta Cryst. D68, 468–477. Web of Science CrossRef CAS IUCr Journals Google Scholar
Moriarty, N. W., GrosseKunstleve, R. W. & Adams, P. D. (2009). Acta Cryst. D65, 1074–1080. Web of Science CrossRef CAS IUCr Journals Google Scholar
Murshudov, G. N., Skubák, P., Lebedev, A. A., Pannu, N. S., Steiner, R. A., Nicholls, R. A., Winn, M. D., Long, F. & Vagin, A. A. (2011). Acta Cryst. D67, 355–367. Web of Science CrossRef CAS IUCr Journals Google Scholar
Negroni, J. (2012). Validation of Crystallographic B Factors and Analysis of Ribosomal Crystal Structures. PhD thesis. EMBL Hamburg, Germany. Google Scholar
Nicholls, R. A., Long, F. & Murshudov, G. N. (2012). Acta Cryst. D68, 404–417. Web of Science CrossRef CAS IUCr Journals Google Scholar
O'Hagan, A. (1994). Kendall's Advanced Theory of Statistics, Vol. 2B. London: Edward Arnold. Google Scholar
Pannu, N. S. & Read, R. J. (1996). Acta Cryst. A52, 659–668. CrossRef CAS Web of Science IUCr Journals Google Scholar
Pozharski, E., Weichenberger, C. X. & Rupp, B. (2013). Acta Cryst. D69, 150–167. Web of Science CrossRef CAS IUCr Journals Google Scholar
R Core Team (2014). R: A Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing. http://www.Rproject.org. Google Scholar
Read, R. J., Adams, P. D., Arendall, W. III, Brunger, A., Emsley, P., Joosten, R., Kleywegt, G., Krissinel, E., Lütteke, T., Otwinowski, Z., Perrakis, A., Richardson, J., Sheffler, W., Smith, J., Tickle, I., Vriend, G. & Zwart, P. (2011). Structure, 19, 1395–1412. CrossRef CAS PubMed Google Scholar
Smart, O. S., Womack, T. O., Flensburg, C., Keller, P., Paciorek, W., Sharff, A., Vonrhein, C. & Bricogne, G. (2012). Acta Cryst. D68, 368–380. Web of Science CrossRef CAS IUCr Journals Google Scholar
Smart, O. S., Womack, T. O., Sharff, A., Flensburg, C., Keller, P., Paciorek, W., Vonrhein, C. & Bricogne, G. (2011). Grade, v.1.1.1. Cambridge: Global Phasing. Google Scholar
Steiner, R. A., Lebedev, A. A. & Murshudov, G. N. (2003). Acta Cryst. D59, 2114–2124. Web of Science CrossRef CAS IUCr Journals Google Scholar
Stuart, A., Ord, K. & Arnold, S. (1999). Kendall's Advanced Theory of Statistics, Vol. 2A. London: Edward Arnold. Google Scholar
Vagin, A. A., Steiner, R. A., Lebedev, A. A., Potterton, L., McNicholas, S., Long, F. & Murshudov, G. N. (2004). Acta Cryst. D60, 2184–2195. Web of Science CrossRef CAS IUCr Journals Google Scholar
Weichenberger, C. X., Pozharski, E. & Rupp, B. (2013). Acta Cryst. F69, 195–200. Web of Science CrossRef CAS IUCr Journals Google Scholar
Winn, M. D., Ballard, C. C., Cowtan, K. D., Dodson, E. J., Emsley, P., Evans, P. R., Keegan, R. M., Krissinel, E. B., Leslie, A. G. W., McCoy, A., McNicholas, S. J., Murshudov, G. N., Pannu, N. S., Potterton, E. A., Powell, H. R., Read, R. J., Vagin, A. & Wilson, K. S. (2011). Acta Cryst. D67, 235–242. Web of Science CrossRef CAS IUCr Journals Google Scholar
This is an openaccess article distributed under the terms of the Creative Commons Attribution (CCBY) Licence, which permits unrestricted use, distribution, and reproduction in any medium, provided the original authors and source are cited.