Models can be constructed for such crystals using TNT, but it requires some work and creativity. And speaking of creative limitations, it is almost impossible to intelligibly describe a solution to this problem in general. I will contrive a very specific case and describe its definition and you will have to extrapolate the principle to your case.
Let us assume a crystal containing dsDNA located on a crystallographic two-fold axis. This DNA is nearly palindromic, with the sequence on one strand of AAATGT and the other ACATTT. Call the first chain W and the second C (Watson and Crick). The molecule crystallizes such that W and C appear superimposed.
You cannot model this situation as a single molecule with statically disordered bases because each conformation consists of a chemically different species. You have to consider this DNA strand as a superposition of two different compounds. From that starting point you will want to apply tricks to reduce the number of parameters to the minimal set, because these crystals usually have a limited resolution.
The crudest model for you situation would be to place both a W and a C strand, superimposed, each with a 50% occupancy. Symmetry will generate the other two (half occupancy) strands and complete the model. This is the most general model and if I had 1.0A data I would consider it.
Since I am at low resolution I want to reduce the number of parameters. How far I can go depends upon the amount of conformational change induced by the difference in sequence. If only the nature and location of the bases are affected I could force the sugar-phosphate backbone to be identical and remove the parameters that allow the backbones to differ. This is likely what I want to do, but I should worry that the backbone does move and watch for it. After all, many papers have been written about sequence specific changes to the backbone.
To reduce the parameters I will have to resort to tricks. What I need to do is to create a new chain, named M, whose sequence file contains monstrous, two-headed, bases at the locations of symmetry breakdown. In the most conservative case I define a single chain with sequence A(AC)AT(TG)T and refine the parameters of eight bases.
Now you probably want to know how to define the monstrous sequence file. I will have to do some geometry definition hacking to pull it off. First start with the definition for the perfect palindrome
(This is just standard DNA.) Now I have to add the second heads to 2 and 5. We put the extra base atoms in residues 2A and 5A and link them to 2 and 5. The link for each is different because one is a purine and the other a pyrimidine and since the atom names differ in each the definitions of the bond lengths and angles must also differ. The new RESIDUE statements will be
Next I have to create definitions for ULINK and YLINK. These links will contain all the geometry definitions you find in $tntdata/nuclgeo.dat that are included in each base that contain some atoms in the base and some atoms in the sugar. If a restraint only contains atoms in the base, ignore it. If a restraint only contains atoms in the sugar, ignore it. Only those restraints which have atoms from both should be put in the link.
Once this has been done I still have a few tasks. All the atoms in 2A and 5A are have 0.5 occupancy and those atoms in the base in 2 and 5 also have 0.5 occupancy. The program must also be told to ignore all bad contacts between the two bases that we know are never really clashing. This is done with the two statements
Of course, if the sugar also changes conformation I have to elaborate this further.