I started looking through code yesterday to figure out how the resnum reference work in the atompair. As a Java programmer I have a built in defense mechanism about trying to tweak C++ code. Reading code does help figure out what is missing in the docs. My first thought was it was the residue number as referenced in the PDB file which means you would need to also give the chain ID which does not appear possible in the atom pair constraints code. It could be as simple as counting residues from 1 to N in the file and using that index. It would be nice if atom pair supported the indexes used in PDB.
I was also thinking that I may play around with the dock_pert option where I create scripts to explore all the rotational and offsets so that instead of nstruct = 1000 you would do nstruct = 1 and submit 1000 low res jobs to the cluster. The only problem is that it looks like the offset is a delta move value for both X,Y where I would need to give a deltaX and a deltaY plus the rotation of the PDB structure to explore the possible patterns for a known interface. Of course I am spending to much time trying to minimize the computational task when I should just submit everything to the cluster as random and let it brute force.
I will probably try and submit 100 jobs where each job does random docking with nstruct=10 to get a 1000 decoys and then move on to the high res option for the best scoring structures assuming they match the interface.