You are here

Relevant number of cycles of loop closure required

2 posts / 0 new
Last post
Relevant number of cycles of loop closure required

Dear All,

I have been working on the Domain insertion module of Rosetta. I need to know what are the relevant number of cycles of loop closure I require to enter for larger domain sets. The domain Im inserting is around 175 aa long. I am having trouble launching the same using MPI in parallel, hence the single processor is taking days for the output. Wondered if anyone who has worked on it could suggest a feasible number for AnchoredPDBcretator and also for AnchoredDesign.


Post Situation: 
Mon, 2018-06-11 02:55

I have no experience with MPI over pyrosetta, but I did write AnchoredDesign.

AnchoredPDBCreator, which creates starting models for AnchoredDesign, does not need very many attempts.  Digging through old notes I see nstruct 500, APDBC_cycles 100000.  I think you can definitely get away with APDBC cycles of maybe 1000-10000 range and nstruct is just "let it run until you have one model good enough for the next step and then kill it".  "Good model" is defined purely on "a well closed loop with no bad bond lengths or angles", torsions are irrelevant.


AnchoredDesign sampling depends on the length of the flexible part of your loop, not the total insertion length.  I assume you have something like 10 flexible residues total.  You aren't designing (which will help), but AnchoredDesign minimizes a lot so it is slow.  If you are ambitious we can hack up the C++ to minimize less frequently - I assume this is a nonstarter since you are using PyRosetta.  I guess you could run with lowered minimization cycles and high tolerance to make it minimize poorly but fast; I don't know what effect that will have on the results; if you want to try that let me know and I will find the flags.  If you are using the -run::min_type flag from the demo options, you can remove it for the newer default minimizer for probably better performance.  I see flag values of 

#-AnchoredDesign:perturb_cycles 5000
#-AnchoredDesign:refine_cycles 10000

left in my notes - those are reasonable starting points.  Nstruct is just "whatever you can afford".  

I won't defend AnchoredDesign as efficent but I will at least try to help you use it!

Mon, 2018-06-11 19:38