You are here

Rosetta relax problem

3 posts / 0 new
Last post
Rosetta relax problem
#1

Hi, I recently got an electron map at 3.8 angstrom resolution. And I have built a protein structure by hand in COOT. Then I refined this structure in PHENIX.real_space_refine package. After refinement, I found that the ramachadran favored percentage was not good. So I followed the simple relax tutorial and did the simple relax refinement to improve the ramachadran. However, during refinement procedure, it appeared that

" core.optimization.Minimizer: (1) WARNING: LBFGS MAX CYCLES 200 EXCEEDED, BUT FUNC NOT CONVERGED! "

How can I modify the parameter to converge the minimization process?

The following is my relax parameter:

mpirun -np 8 $ROSETTA3/main/source/bin/rosetta_scripts.linuxgccrelease \
  -database $ROSETTA3/main/database/ \
  -in::file::s 1A_1F_ref3fit.pdb \
  -parser::protocol realx.xml \
  -ignore_unrecognized_res \
  -edensity::mapreso 3.8 \
  -edensity::cryoem_scatterers \
  -crystal_refine \
  -out::suffix _relax \

  -default_max_cycles 200

The xml file is :

<ROSETTASCRIPTS>
        <SCOREFXNS>
                <dens weights=talaris2013_cart>
                        <Reweight scoretype=elec_dens_fast weight=25.0/>
                </dens>
        </SCOREFXNS>

        <MOVERS>
                <SetupForDensityScoring name=setupdens/>
                <LoadDensityMap name=loaddens mapfile="1A_F.mrc"/>
                <FastRelax name=relaxcart scorefxn=dens repeats=1 cartesian=1/>
        </MOVERS>

        <PROTOCOLS>
                <Add mover=setupdens/>
                <Add mover=loaddens/>
                <Add mover=relaxcart/>
        </PROTOCOLS>

        <OUTPUT scorefxn=dens/>
</ROSETTASCRIPTS>

 

 

 

Category: 
Post Situation: 
Wed, 2016-03-09 21:41
jingwei xu

The "WARNING: LBFGS MAX CYCLES 200 EXCEEDED, BUT FUNC NOT CONVERGED! " message is somewhat common, and nothing to be too concerned about. Generally speaking, you'll go through multiple rounds of modification and minimization, and even if you don't converge on earlier rounds, you should refine things further in subsequent rounds. It's a cost/benefit trade-off: you could run more cycles, but it will take more time and not result in substantially better structures - 200 cycles gets "close enough".

The key thing is to examine the output structure and make sure that the the structures look good.

Fri, 2016-04-29 08:41
rmoretti

Thanks for your help. I have refined my structures several times, and it indeed helped to improve my structure.

Best wishes.

Wed, 2016-05-04 23:27
jingwei xu