I want to locally dock a 10 kDa protein A in the middle of a 4-subunit protein assembly (BCDE, ~50 kDa each). I roughly know the orientation of A on the BCDE platform, where A contacts every other chain. After placing A near its target binding site I first run relax -nstruct 5 and use the 'best' model for a subsequent full local docking (A_BCDE) with rosetta scripts (linuxgccrelease v3.7). I am getting reasonable energy funnels (both I_sc and total score vs. RMSD) with -nstruct 50, but I wonder if my sampling scale is appropriate for the task.
Could the approach described above be considered correct and meaningful? I've seen examples with 10^5 decoys, but not sure if that's feasible in my case.
If increased sampling is absolutely essential for the given DOF, would it be generally recommended to use an online server (Docking2 on ROSIE? Or FunHunt? - What would be the number of considered decoys in such a black box?) or gain access to a CPU cluster at a host institution, if available?
Any advice will be greatly appreciated.
With thanks and best regards,