I found that almost all the docking tools including Rosseta try to use the entire carbohydrate molecules for docking, but in fact from many X-ray crystal stuctures we found that if a carbohydrate molecules contains three or four ring, it usually just use the one terminal ring to form some H-bond with the protein. For such case, Glide or other tools would very difficult to get a reliable docking results.
Rosetta coupled with other docking tools tend to dock the oligosaccride with full molecule, does this mean that current docking tools emphasize too much on enthalpy calculation while the entropy calculation is inefficient? In this case, leaving some rings of oligosaccride in solvent would be considered unfavorable by the docking tools. How do you think of this?
I would be very appreciated if someone could give me some advices on how to avoid Rosseta use the entire carbohydrate for docking and how to get a much more reliable results for carbohydrate docking?
Thank you very much
If I understand correctly, you want Rosetta to produce models with the entire carbohydrate present, but only use part of the carbohydrate for docking, and ensure the remainder is free in solution?
The scorefunction tends to be very agglutinating, as a general rule. This makes sense due to its roots as a protein-folding algorithm, which explicitly requires more and more interactions.
Rosetta doesn't have any easy solutions for ignoring parts of molecules, or treating different parts differently. One way to accomplish this would be to modify the atom types on the extra carbohydrate rings. You could try altering all the extra carbohydrate atoms to virtual atoms - this will make them invisible to the scorefunction (of course, it will also mean they don't clash anymore, so who knows where they'll end up.)
You could try creating new faux atom-types with solvation parameters that are off the scale. Edit database/chemical/atom_type_sets/fa_standard/atom_properties.txt to clone the atom types represented by your carbohydrate, then alter the solvation parameters of the clones (the LK columns) to make them impossible to desolvate. This should forcibly keep them in solution. Edit your parameter file for the carbohydrate to use the new faux atom types and see what happens.
Thanks a lot for your kind reply.
I don't mean to use Rosetta to produce models with the entire carbohydrate present, but only use part of the carbohydrate for docking.
What I want to do is use Roseeta to judge and dock correctly carbohydrate by itself. Because, if there is no crystal structure, I don't know which part of the carbohydrate would have H-bond with protein, so neither which part will not have H-bond with protein.
So, I am wondering is it possible for Roseeta make such kind of decision by itself? Or should I combine other kind of methods for further validate such as: MD?
Rosetta's score function is probably too greedy (or perhaps just insufficiently aware of entropy) for it to dock only part of a sugar the way you are suggesting.
You can try playing with scorefunction weights - if you weaken the long-range hydrogen bond terms, then perhaps Rosetta will make only the strongest (most real) hydrogen bonds to your carbohydrate.
You could also do the experiment I described iteratively. Try it with each different part of the molecule as "real" and the others virtualized, and see which set of models has the best scores - that's Rosetta's prediction.