BioCrystallographica,

a package for doing Crystallography with Mathematica

 

Nicolas Ambert*, Julien Vanwinsberghe*, Philippe Dumas.

 

Equipe de Cristallographie

UPR9002 du CNRS conventionnée avec l’ULP

IBMC, 15 rue René Descartes 67084 Strasbourg cedex FRANCE

 

* N.A & J.V contributed equally to this work

 

Corresponding author: P. Dumas

e-mail: p.dumas@ibmc.u-strasbg.fr

 

 

 

Abstract

 

We have made a package for doing macromolecular crystallography with the software Mathematica from Wolfram Research. This package contains in its present state all necessary functions for handling files commonly used and performing all basic calculations required in daily work. Our goal was obviously not to offer a stand-alone alternative to well known (and huge) crystallographic packages like CCP4 or CNS, but to open the way for using a programming language with very rich symbolic capabilities. To take benefit of existing softwares, we have made possible to call various CCP4 routines from within the package and to get the results immediately available in the package environment. Such possibilities could be easily extended to CNS, or other packages. In its present state, BioCrystallographica is not really intended for a common usage, but rather to allow those involved in the field to analyze their results and develop new methods. The Mathematica  programming language, as other languages of that kind, requires a significant practice before being used efficiently, particularly for abandoning procedural programming methods in favor of functional ones. This investment, however, is worth the effort in view of the extreme density of the code and of the depth of the questions that can be addressed.

 

 

1. Introduction

 

There now exists a large number of crystallographic softwares, most often organized in more or less general packages. CCP4 (CCP4, 1994) and CNS (Brünger et al., 1998) are particularly well known examples. Having a look to Jean Cavarelli’s web site in Strasbourg provides an almost exhaustive list of available softwares. All of them rely on different kinds of programming languages, from the venerable Fortran  to the ‘up-to-date’ Java and Python, passing by the not so young C++. To our knowledge, no serious attempt of using a programming language with symbolic capabilities like Mathematica has been made. Several reasons can be put forward to explain this situation. Probably, the most important ones are (i) that such a programming language is considered as slow and, (ii) that it is seen as expensive. However, we think that such a tool offers invaluable possibilities that cannot be found in any existing crystallographic softwares. Indeed, although packages like CCP4 and CNS are extremely versatile by allowing to combine at will individual programs, they do not allow to perform immediately an original methodological test requiring a not yet programmed function. On the contrary, the environment offered by a programming language like Mathematica authorizes to make such a test very easily.

 

 

2. Essential features and working environment of Mathematica

 

Mathematica is organized in two parts: the so-called ‘front-end’ corresponding to the program-user interface, and the ‘kernel’ receiving the commands from the front-end, doing the calculations and returning the result to the front-end. Although it is still possible to interact directly with the kernel in an ‘old-fashioned’ way (i.e. at the command-line level, e.g. in a unix window), it is obviously much more efficient and user-friendly to do so through the front-end. This allows to obtain directly graphical results immediately after the corresponding lines of code, and also to interact very efficiently with an exhaustive on-line help. The front-end interface is organized by the user into individual cells that represent a logical unit of code. This organization is flexible because it may be modified at will. The input code is interpreted cell by cell if desired, which also provides a high degree of flexibility during development.

 

It should be realized that the essential characteristics of a software like Mathematica resides in its symbolic capabilities. Therefore, it differs from softwares devoted to numerical calculations by the fact that, by default, it does not assume that a general symbol has a precise meaning. For example, defining a simple function like  without mentioning anything about the parameters a, b, c and the variable x lets open the possibility that these may correspond to usual real numbers (with finite machine precision), to numbers corresponding to symbols with infinite precision (e.g. p ), to complex numbers, to matrices, or even to yet undefined symbols that will allow such algebraic calculation. Obviously, such a burden may have a price in terms of speed, but may also yield results unattainable with classical ‘purely numerical’ softwares. However, when restricting the input data to numbers with finite precision, the calculations become comparable in execution time to that obtained with more specialized softwares. For example, obtaining the eigen values of a large numerical matrix is in now way particularly slow.

 

3. Illustration of the symbolic capabilities

 

Suppose we are interested in expressing the general form of a matrix representing a rotation of angle  around an axis defined by a unit vector u. Of course, such simple objects are immediately available in the set of predefined tools. However, we will show as an example how to define and use it symbolically. For avoiding lengthy developments beyond the scope of this note, several technical aspects will not be detailed, or will even be left in the shadow, when the context is sufficiently clear or when this does not compromise understanding. Everyone interested in details should search in the on-line help for additional explanations (and accept to spend some learning time).

 

We can first make use of the intrinsic definition of a rotation.  Rotating a vector V by q around a unit vector u defining the axis of rotation may be expressed as :

 

                                                     (1)

Such an expression has an obvious geometric meaning in terms of the component of V along the axis of rotation u (invariant under the rotation),  and of the rotated component of V perpendicular to u. Equation (1) has an immediate translation in the Mathematica language:

 

RotVec[u_List,q _][V_List]:=(1-Cos[q ])*(u.V)*u + Cos[q ]*V + Sin[q ]*Cross[u,V]

                                                                                                                                             (2)

 

This defines a function Rotvec depending on the two parameters u and q and taking V as the variable. Note that apart for idiosyncratic specificities inherent to any programming language, the latter definition (2) is easily decipherable. The nice thing is that (2) can be used symbolically with u defined as {a , b , g } and V as {x, y, z} without giving explicit values to the variables. As detailed a little bit more in the following, the braces { } are the hallmark of any kind of list in Mathematica and a vector may be considered as a 1D list. One may thus ask for the result of :

 

W = RotVec[{a , b , g },q ][{x, y, z}]                                                (3)

 

which will appear as a list of three linear forms in x, y and z. By applying the following syntax (which will not be explained in details here, but see § 6),

 

             m = Map[Coefficient[#,{x,y,z}]&,W]                                                        (4)

 

the coefficients of x, y and z are obtained, which after an additional line of code for factorization (not shown here), yields the sought after rotation matrix  m :

 

(5)

                                                                                                                                 

Note that the latter expression was not hand-written, but is a direct output of Mathematica. In order to achieve giving a glimpse on the symbolic capabilities, we will simply show the result of the function Eigenvalues  applied to the latter matrix m:

 

Eigenvalues[m] /.( a ^2+b ^2+g ^2)® 1                                                                       (6)

 

In (6), the syntax   /.(a ^2+b ^2+g ^2)® 1 is a ‘replacement rule’ forcing u to bea unit vector. As output of  (6), one readily obtains the list of eigen values:

 

            {1, Cos[q ]-iSin[q ], Cos[q ]+iSin[q ]}                                            (7)

 

This is a useful reminder that  and , as well as 1, are eigen values of a matrix of rotation by an angle q . Asking for the corresponding eigen vectors with:

 

FullSimplify[Eigenvectors[m]] /.( a ^2+b ^2+g ^2)® 1                                  (8)

 

one obtains as output:

   (9)

 

containing, first the expected vector {a /g , b /g , 1} parallel to u ={a , b , g }, as well as those corresponding to the complex eigen values  and . In (9), FullSimplify is a call to a simplification procedure of wide use for algebraic expressions.

 

 

4. Description of the basic set of utilities available in the package

 

We will only give a brief list of essential utilities.

 

4.1 Reading (and making) files

PDB coordinate and reflection files

Denzo and CCP4 MTZ reflection files (MTZ files can be read and written)

CNS and CCP4 electron density maps (CCP4 maps can be read and written)

 

4.2 Atomic model representation

A simple graphics representation of an atomic model is available. For now, this is only a very crude option for the sake of verification of the coordinates in use.

 

4.3 Crystallographic data bases (from CCP4)

All symmetry operations.

Limits of asymmetric units and alternative origins for each space group.

All atomic scattering factors.

 

4.4 Elementary calculations

Orthogonalization and deorthogonalization matrices.

Metric tensor in reciprocal space, calculation of resolution.

Sorting out centric and  general reflections.

 

4.5 Phase and map calculation

Phase and map calculations can be made either internally (i.e. as a Mathematica procedure), or by launching from within Mathematica in a ‘user-transparent’ way a CCP4 calculation (i.e. without any need of typing a DOS or Unix line of command or of using the CCP4 GUI).

 

4.6 Plotting and graphics representation

Mathematica provides a great deal about plotting data and graphics representation. There exists many possibilities for exporting and importing graphics. This aspect is not secondary because working in such an environment makes graphics representation immediately available for anything. Map section contouring, for example, is extremely efficient and fast. In the present state, however, it is not possible to make plots with oblique axis as requested for non orthorhombic space groups. It is intended to remove this limitation in the next release of the package.

 

4.7 On-line help

It is an important feature that all tools in the package are referenced in the on-line help, exactly as for any ‘official’ Mathematica tool. It should also be emphasized that this on-line help is not only an exhaustive list of the syntactic rules, but that it also allows a dynamic test of them. For example, when searching for how a call to CCP4 is made for calculating an electron density map, the user has the possibility, within the help itself, of running real examples and changing parameters.

 

 

5. Organization of data

 

One major tool in Mathematica is the ‘list’. As previously mentioned, a ‘list’ is simply notated by braces enclosing different elements separated with commas (e.g. FirstPrimes = {1,2,3,5,7}; one particular element, for example 5 at the fourth position, is retrieved as FirstPrimes[[4]]). It can be viewed as representing any collection of any objects structured in any possible way. A list can be as simple as a 1D vector of numbers (atomic coordinates are obviously structured as a list {x,y,z}), but can collect symbols, e.g. , or even symbolic rules. It can be structured to any depth, regularly as a m´ n matrix, or without any regularity. It is of such use that many powerful tools exist to manipulate them. Simple examples will illustrate its usage.

 

5.1 Definition of ‘parallel lists’

What we call ‘parallel lists’ is of extremely common usage in the package. Two lists are said ‘parallel’ if each element in one list corresponds to the element at the same position in the other list. As a consequence, ‘parallel lists’ must have the same length. The importance of this notion is made very clear in the following with the organization of atomic coordinates or of reflections. For example atomic coordinates and the corresponding atomic names are stored in ‘parallel lists’. Importantly, the substructure of each list may be different; one only requires that the two lists be identically structured at the first level.

 

One very important function allowing to make a single list from two parallel lists or, alternatively, to split a single list into parallel lists is TRANSPOSE . For example, with the two parallel lists:

 

hkl = {{h1,k1,l1},{h2,k2,l2},{h3,k3,l3},…};

 

F  =  {F1, F2, F3,…};

 

one obtains a single list:

 

hklF = {{{h1,k1,l1},F1},{{h2,k2,l2},F2},{{h3,k3,l3},F3},…};

 

with the syntax:

 

            hklF = Transpose[{hkl,F}];

 

Conversely, hklF may be splitted into hkl and F with the syntax:

 

            hkl = Transpose[hklF][[1]];

F   = Transpose[hklF][[2]];

 

Note that the semicolon at the end of each definition tells Mathematica not typing the result on the screen (which may correspond to huge outputs with long lists).

 

5.2 Atomic coordinate organization

Atomic coordinates  are naturally organized as a list with the hierarchy :

 

{chain(s) {residues {atoms}}}.

 

If a list of coordinate is obtained from a PDB file, by default it is named XYZ, and XYZ[[2]] will refer to the second chain, XYZ[[2,3]] to its third residue, and XYZ[[2,3,1]] to the first atom of the third residue of the second chain. XYZ[[2,3,1]] thus corresponds to a list of atomic coordinates of type {X,Y,Z}. All other relevant pieces of information for each atom are stored in different ‘parallel lists’. For example the list named QB stores occupancies Q and temperature factors B, and QB[[2,3,1]] refers to the two-value list {Q,B} for the atom with coordinates XYZ[[2,3,1]]. The same holds true for the lists ATOMNAMES and ATOMTYPES. For the sake of consistency and programming generality the same structure is always used, even if one level is not useful in a particular case. For example, for a molecule with two different chains the coordinates are naturally stored as {{chain1},{chain2}}, but for a molecule with one single chain they are still kept as {{chain1}}, not {chain1}.

 

5.3 Symmetry operation organization

All space groups are stored in a list of sublists, with one sublist for each space group. The necessary information was obtained from the CCP4 package. The information for a specific space group is retrieved as:

 

            InfoSpaceGroup["P21"]  (or InfoSpaceGroup[4])                 (10)

 

which yields as output the following list:

 

                                 (11)

 

In the fourth position are stored the symmetry matrices and in the fifth position the corresponding translations.

 

5.4 Reflection organization

Reflections are stored in ‘parallel lists’. One of these lists, commonly named hkl, stores Miller indeces as a list of individual {h,k,l}. Intensities, amplitudes, sigmas, phases, etc… are stored at will of the user in ‘parallel lists’. Note that the user is free of grouping whatever he wants in a single list. For example, if F’s and sigmas are grouped to form FSigma, FSigmais madeof elements {F(h,k,l),SigmaF(h,k,l)}. The two lists hklandFSigmaare obviously parallel.

 

5.5 Electron density maps

Electron density maps in 3D are stored in lists of sections, each section being a list of rows, and each row a list of values. They can be readily obtained as such by a call to CCP4 from within Mathematica:

 

DensityMap = CalcMapCCP4[hkl,F,PHI,UnitCell,SpaceGroup,(Options)](12)

 

where hkl is a list of Miller indeces,  and  F and PHI the parallel lists of structure factor amplitudes and phases. UnitCellis a list of the form {a, b, c, a , b , g } andSpaceGroupa character string of the form “P21”. All options for CCP4 map calculation can be given in a list of Options.

 

 

 

 

 

6. Examples of ‘elementary’ manipulations

 

A few examples will be worked out in some details to illustrate the usage of the stored data, as well as some basic features of the philosophy of Mathematica programming, and the density of the code. As already stated before, several technical aspects will not be developed, or will be left in the shadow, when the context is sufficiently clear or when this does not compromise understanding.

 

6.1 Manipulating electron density maps

Manipulating an electron density map as a whole is a trivial operation in Mathematica. For example, applying a mask mask onto a density map map  for solvent flattening is obtained as:

 

FlatMap =  mask*map ;                                                                                              (13)

 

Predefined interpolation tools allow to determine functional approximations of a map very efficiently. This allows, for example, to resample a map at will on any ‘skewed’ mesh for molecular replacement. Such tools seem promising, also in the scope of molecular replacement, for applying smooth distortions on a map by using normal-mode-like methods as already described (Delarue & Dumas, 2004; Suhre & Sanejouand, 2004). We think that there exist considerable possibilities in this area because of the great efficiency of many preexisting tools in Mathematica. This will not be detailed further here.

 

6.2 Calculating scattering factors

The scattering factor of a particular atom type is obtained as InfoFormFactor[“atomtype”], “atomtype” representing a character string. For example, InfoFormFactor[“Ca”] returns the information for calcium:

 

{Ca,{{8.6266,10.4421},{7.3873,0.6599},{1.5899,85.7484},{1.0211,178.437},{1.3751,0.}},{0.341,1.286,0.203,0.306}}

 

where the parameters for the usual four-gaussian approximation (International Tables, Vol 4, 1974) are returned as the second element of the latter list, that is to say:

 

  GaussCa = InfoFormFactor[“Ca”][[2]];                                                           (14)

 

which contains the following numerica values:

 

{{8.6266,10.4421},{7.3873,0.6599},{1.5899,85.7484},{1.0211,178.437},{1.3751,0.}}

 

Summing up the five functions (with ), corresponding to the five elements of form {Qi, Bi} in the latter list GaussCa (note that the fifth Gaussian function is in fact a constant), can be performed with the self-explanatory usual ‘procedural’ syntax:

 

Sum[g[GaussCa[[i,1]], GaussCa[[i,2]], s], {i, 1, 5}]             (15)

 

Although it appears quite cryptic at first sight, it is much more preferable to use the following functional method. First, one ‘maps’ the function g onto the list GaussCa, that is onto each element of GaussCa:

 

            Map[g[#[[1]],#[[2]],s]&, GaussCa]                                                       (16)

which yields as result the following list:

 

{Q1*Exp[-B1 s^2], Q2*Exp[-B2 s^2], Q3*Exp[-B3 s^2], …}        (17)

 

with Q1=8.6266, B1=10.4421, Q2=7.3873, B2=0.6599,etc…In (16) the syntax #[[1]] and #[[2]] stand for the arguments of a function (here, the function g) by referring to each element of the list it is mapped onto. In the present case, each element of GaussCa is a sublist of the kind {Qii, Bi}, and #[[1]] thus refers to Qi, and #[[2]] to Bi. Note also the character ‘&’ that is the mark of the end of the definition of what is called a ‘pure function’ in Mathematica. The usage of # for referring to variables, and of ‘&’ for marking a pure function cannot be dissociated. This syntax allows using a function without predefining it. In some sense, a pure function may be viewed as a ‘disposable’ function. For example, the sum of the five functions  can readily be invoked as follows (note the use of the trace operator Tr for summing up the elements of a 1D list):

 

diffus[GaussX_List][s_]:=Tr[Map[#[[1]]*Exp[-#[[2]]*s^2]&,GaussX]         (18)

 

to define the value of the scattering factor of any element X with parameters stored in GaussX. The form used in (18), i.e. the mapping of a pure function onto a list, is equivalent to that used in (15) as far as the result is concerned, but (18) has the advantage of yielding very often the most concise code, and of never requesting explicit handling of the limits of a list. Since Mathematica has symbolic capabilities, GaussX may be replaced with a list of symbols like {{Q1,B1},{Q2,B2},{Q3,B3},{Q4,B4},{c,0}} to obtain a purely analytic expression.

 

6.3 Sorting out centric and general reflections

As another example, we will consider a concise way (and even also a very concise way) of extracting all centric reflections from a list of reflections. A reflection {h,k,l} is centric if there exists a symmetry operation whose matrix m is such that :

 

{h,k,l} + Transpose[m].{h,k,l}  = {0,0,0}                                     (19)

 

One thus first maps the pure function (# + Transpose[m].# )&onto  the full list of reflections hkl and then searches for positions of {0,0,0} in the result; this can be made a function of m and hkl as follows:

 

  CenPos[m_List,hkl_List]:=Position[Map[#+Transpose[m].#&,hkl],{0,0,0}](20)

 

Then, one can map this new function onto the list Sym of all symmetry matrices for the space group of interest (see § 5.3), and consider the union of the whole set of positions:

 

            CenPositions = Apply[Union , Map[CenPos[#, hkl]&, Sym]];

 

From this list of positions in hkl, one can then extract all centric reflections from hkl:

 

            Centric = Map[Extract[hkl,#]&, CenPositions];

 

In the package, these three lines of code are grouped into a callable function (named CentricHKL), and the list of centric reflections is returned at once as:

 

            Centric = CentricHKL[hkl,SpaceGroup];

 

In fact, these three lines of code can even be reduced to a bit cryptic ‘one-liner’:

 

Union@@Table[Cases[hkl,_?(#+Transpose[Sym[[i]]].#=={0,0,0}&)],{i,2,Length[Sym]}]

(21)     

This is only shown here to illustrate how concise the code can be and this will not be discussed further. The former three-line version is used in the package because it turns out to be faster. This example is only a very short glimpse at the powerful capabilities of Mathematica for sorting and extracting information from any kind of lists.

 

 

7. Conclusion

 

In its present state, BioCrystallographica allows performing original methodological tests because all basic tools for doing crystallography are available and embedded in a very powerful environment. We hope a significant number of persons among those interested in methods will participate in developing it. We consider as an interesting goal the extension of  links between BioCrystallographica and other crystallographic softwares as this is already done with CCP4. Clearly, reinventing the wheel and reprogramming already available tools has no interest, but making them available in an original powerful environment is of great interest. For the time being the link to external programs (like CCP4) has been tested under Windows. Extending this functionality to Unix-, or Linux-based platforms should be straightforward.

The package is freely available for academic users upon request. We have made a web site to illustrate our work with Mathematica in crystallography. For the time being, it shows essentially teaching applications. We plan to enhance the functionalities in a near future.

 

N. Ambert (2003) and J. Vanwinsberghe (2006) made each a four-month student work.

 

References

COLLABORATIVE COMPUTATIONAL PROJECT, NUMBER 4. ‘The CCP4 Suite: Programs for Protein Crystallography’. (1994) Acta Cryst. D50, 760-763.

Brünger, A.T., Adams , P.D., Clore, G.M., DeLabo, W.L., Gros, P., Grosse-Kunstleve, R.W., Jiang, J.-S., Kuszewski, J., Nilges, M., Pannu , N.S. , Read, R.J., Rice, L.M., Simonson, T. & Warren, G.L. (1998). Acta Cryst. D54, 905-921.

 

Delarue, M. & Dumas, P. ‘On the use of low-frequency normal modes to enforce collective movements in refining macromolecular structural models’. (2004) PNAS 101, 6957-6962.

 

International Tables for X-ray crystallography, Vol 4 (1974) The Kynoch press, Birmingham , pp. 99-101.

 

Suhre, K & Sanejouand Y.H. ‘On the potential of normal mode analysis for solving difficult molecular replacement problems’, (2004) Acta Cryst. D60, 796-799.