Theres a topic?
X-ray...cool. Been a while since I did that, and it was only crystalography. Probably much more straighforward.
uhm... cristallography is done with x-rays, so i think it's the same thing...
yeah, I know its the same method - its that the x-ray diffraction shows clear repeating geometries in crystalography given different lattice shapes, and its pretty easy to interpret.
Organic molecules would be harder.
Actually, organic small-molecule crystallography is considerably easier. Due to the nature of the beast, those guys tend to grow crystals on the scale of sugar granules in granulated sugar, and those crystals tend to diffract very strongly, giving high resolution (usually considerably better than 1 Angstrom). In protein crystallography we tend to deal with crystals in the order of 10-50 micrometres in each dimension, which in general you have to 'fish' by hand from a drop of anywhere from 4 microlitres to 100 nanolitres with a small nylon loop, mounted on the end of a metal pin. Once you have the crystal in the loop (probably having first transferred it using the same loop method to another drop which contains a cryoprotectant solution to stop ice formation in the next step), you then typically plunge freeze the crystal in liquid nitrogen, or place it in a cryostream of nitrogen gas at 100K.
The next tricky thing is the sheer number of approximations required in protein crystallography. Assuming that you have a protein crystal which diffracts X-rays, and that those diffracted X-rays are reliably measurable to a usable resolution (we tend to like 2.5 Angstroms or better for the ligand-bound structural work I do), you then integrate your whole dataset (something like 50-200 images, each around 10-20 megabytes), and solve the structure (increasingly by using an existing structure that's similar to generate an approximate set of phases (information which you cannot gain from the diffraction pattern), but also by locating heavy (or anomalously dispersing) atoms in the unit cell (exactly like small molecule crystallography)).
Having solved the structure, you then need to refine it. You'll typically have 1000-4000 heavy (non-H) atoms in the single protein molecule, so the positions, temperature factors and occupancies of these atoms need to be refined - the product of which is your 'X-ray crystal structure', which is in fact a model based on the data you collected.
Because the data you collected probably comprises something like 10,000 to 30,000 unique data points, and you have, say 2000 atoms, each with x, y and z coordinates (that's 6000 parameters), and some contribution from temperature factor estimation, and probably a further 20-400 water molecules (depending on resolution), each of which has another 4 parameters associated (assuming an occupancy of 1 for each), your observation to parameter ratio gets quite small (approaching 1 is bad!).
Protein crystals which diffract more strongly and give measurable diffracted X-ray spots at around, say, 1.5 Angstrom resolution tend to give much more reliable models, but even these require a LOT of approximations. We are helped enormously by the requirements of chemical bonding, of course: peptide bonds tend to be planar, due to the lone pair of electrons on amide nitrogen atoms being partially delocalised with the formal pi orbitals of the adjacent carbonyl group; bond angles and bond distances tend to be pretty well defined; there are fairly few rotamers (conformations of amino acid side-chains) which are much more likely than others (again, due to chemical bonding restraints); certain values of phi and psi (and omega) - the angles within and between adjacent peptide bonds - are much more likely than others, and define the secondary structural elements, such as alpha helices, beta sheets, strict beta turns, eta helices and so on. All of these restraints, which we know largely from small molecule crystallography, help to reduce the apparent observation to parameter ratio by restricting the values that some of our parameters can take (eg by limiting in space (x,y and z) where a certain atom can be based on its bonding pattern with neighbouring atoms).
Oops, looks like I got a bit carried away and re-derailed this topic again...
Roo
