I seeing tens of thousands of PDB files on the internet. I really want to determine a 3D structure of my protein of interest. I've heard that 3D structure determination is a complex, expensive, and specialized procedure that can take months or years and is hard to perform routinely or even purchase commercially.
Could you please explain why and how hard it is? What are the requirements and how big of a budget is typically needed to perform this kind of experiment?
Experimental protein structure determination is hard: the most common method is X-ray crystallography, which can be done in a few months if you are lucky and can take years if you're not. The problem with X-ray crystallography is that you need good protein crystals, and in most cases, proteins don't crystallize very well, so it takes a lot of time (and a lot of purified protein) to get the right crystal. (In the comments section Cody added some good links with more in-depth abut X-ray crystallography for structure determination relating to ATP synthase).
The cost is mainly based on the materials & manpower you need for this, and you need to test a lot of different solutions to crystallize your protein in, which aren't very cheap. If you don't have access to specialised equipment for making the crystals (mostly pipetting robots) or the measurement (X-ray beamline), this is going to be super expensive.
Crystallography requires protein crystals, which are sometimes too troublesome to obtain. Other methods that do not require crystals, such as NMR and cryoEM, are even more difficult to perform though less luck dependent. These techniques rely on very expensive equipment and often cannot resolve the structure as precisely as protein crystallography. So unless you can find one of the few people who have experience with these methods, you'll run into a lot of additional problems.
There is also computational structure prediction, which only needs a powerful computer. However, unless your protein is very similar to one with a known structure it will probably not work reliably. As mentioned in the comments there are web server's for the established methods and progress is constantly made on new(er) algorithms, so depending on your needs it's definitely worth a shot to try computational structure prediction.
I'll address NMR for structure determination. It is the less common method, only ~10% of protein structures are determined this way, though it has e.g. advantages for nucleic acids and more than a third of those was solved by NMR. Take any numbers here as very rough estimates, there are a lot of factors that influence the difficulty and cost.
For NMR, the single most important factor is size. A stable, well-behaved protein of 10-20 kD is pretty much routine. Large proteins are either very difficult to measure or outright impossible by NMR.
You need large quantities of protein for X-ray and NMR, but in the case of NMR you also need it isotope-labeled. Uniformly labelling proteins 15N/13C isn't all that expensive if you can express your protein in E. coli in minimal medium (around $100 per liter medium or so, mostly in 13C glucose), but it can get really expensive very fast if you need full medium or anything fancy.
Your protein needs to be stable in solution for a few days at least, a single 3D experiment can take multiple days to measure. If your protein is barely stable enough, you also have to produce much more protein because you need to use many samples to perform all necessary experiments. A complication here is that you can't add a lot of salt to an NMR sample without drastically reducing the sensitivity of the NMR experiments. But some proteins are hard to get stable in a low-salt environment.
You need something like a few weeks of measurement time on a high-field NMR spectrometer, though this depends a lot on the size of the protein and the concentration you can achieve in your sample and the general difficulty of assigning the protein resonances. This kind of spectrometer can cost a few millions, and you need to supply liquid nitrogen and liquid helium to keep them working. Something like a $1000 per day measurement time is a number I've heard about the cost, but there are a lot of factors that go into this and I can't say how accurate this number is.
Then you need to assign the resonances in your protein, which takes a single person several months. This again varies a lot depending on the difficulty.
Then you need to calculate the structure, and typically this requires several rounds of refining the assignment and analysis and redoing the calculations. It's not really expensive in terms of computer power, but it takes quite some work by the person running the calculations and checking them.
