I normally use the ADT GUI, but it's quite slow, so I just tried the command line protocol that Oliver suggested. It works, but I recommend leaving off the '-Z' option, which restrains
all of the torsions (including hydroxyl orientations). By default, all rotatable bonds are active except for amide and guanidium.
Autodock Vina (and therefore Vina-Carb) does not include a scoring term for electrostatics, simply relying on a hydrogen bonding term. As a result, the program has a difficult time
with blind docking, or choosing where the binding site should be on the protein surface. However, if you already have a good idea of where the ligand binds (i.e. alanine scanning, near the catalytic residue, etc.), Vina-Carb can produce accurate models. It
helps if you have a positive control.
For example, I successfully docked the substrate within the co-crystal of an enzyme (PDB ID: 4NDZ) back into the protein with an RMSD of the recognized trisaccharide (GlcNS-IdoA-GlcNS) of < 1 Angstrom. This gave me confidence in using the program on the same
enzyme, with the same settings, in order to predict the binding mode of alternative ligands.
Due to the large number of rotatable bonds, as well as the size of the grid box that is often required, you should try increasing the exhaustiveness within the config file. The default
is 8, so I used 80 in the example described above.
Please let me know if you have any other questions.