Dear Amber users!
Here are some updates of my tests.
basically I reduced size of my system using monomer protein docked to
the same ligand and run MMGBSA.py.MPI (using 20 processors) on 120
frames (45 ns) with normal mode calculations, which did the job during
1 day. Since I still have the problem with the same computing using
dimeric protein, would it be reasonable just to strip a part of the
protein not involved in the binding (another monomer) during the
post-processing analysis ? If such trick is possible, what should be
the proper mask for ante-MMPBSA.py ?
Many thanks in advance
Enrico
Il giorno ven 12 ago 2022 alle ore 09:16 Enrico Martinez
<jmsstarlight.gmail.com> ha scritto:
>
> I've just double checked.
> With the quasi-harmonic approximation the calculation took 1 day for 5
> frames and the result for dH is the same as it was with 300
> snapshots. When I tried using 5 threads with 50 snapshots I got the
> error after 4-5 hours of computing..
> So what should I do now? :-)
>
> Il giorno gio 11 ago 2022 alle ore 17:27 Bill Miller III
> <brmilleriii.gmail.com> ha scritto:
> >
> > If normal mode analysis will not work on 4 frames (if it crashed after only two hours, it probably ran out of memory during calculation of the first frame) on your computer, then I would suggest using a computer with more RAM available. If that is not an option, then I should try to use the quasi-harmonic method on as many frames as your computer's resources will allow.
> >
> > -Bill
> >
> > On Thu, Aug 11, 2022 at 10:11 AM Enrico Martinez <jmsstarlight.gmail.com> wrote:
> >>
> >> and for my system of dimeric protein (complex w/o solvent is about 9k
> >> atoms) what options of the MMBSA with entropy computing would work?
> >> How many frames may I use with the parallel version of mmpbsa.py.MPI?
> >> What options for the NMA?
> >> I've already tried with
> >> nmstartframe=1, nminterval=4, nmode_igb=1, nmode_istrng=0.1,
> >> #maxcyc=20000, drms=0.0001,
> >>
> >> and eventually it crashed after 2h of computing.. :-)
> >>
> >> Il giorno gio 11 ago 2022 alle ore 17:06 Bill Miller III
> >>
> >> <brmilleriii.gmail.com> ha scritto:
> >> >
> >> > The calculation of entropy uses the resources differently. :)
> >> >
> >> > Normal mode will only perform the calculation one frame at a time, so it is MUCH slower than the quasi-harmonic method and therefore users typically analyze significantly fewer frames with normal mode analysis than quasi-harmonic. So normal mode analysis will use less resources (e.g. RAM) if you have a much smaller system (regardless of the number of frames), but becomes exponentially more challenging with a larger system.
> >> >
> >> > However, I think the literature would suggest that normal mode calculations are oftentimes combined with MMPB(GB)SA energies more often than quasi-harmonic entropy calculations are.
> >> >
> >> > Hope that helps.
> >> >
> >> > -Bill
> >> >
> >> > On Thu, Aug 11, 2022 at 10:02 AM Enrico Martinez <jmsstarlight.gmail.com> wrote:
> >> >>
> >> >> Thanks for this info Bill!
> >> >> It seems to me that the problem might be due to the RAM
> >> >> I am on Dell workstation equipped by 96 CPUs, here is the info regarding RAM:
> >> >> Mem: 125Gi 60Gi 23Gi 297Mi 41Gi 64Gi
> >> >> Swap: 2.0Gi 2.0Gi 10Mi
> >> >>
> >> >> Does the calculation of entropy using normal mode analysis require
> >> >> less resources ?
> >> >> Thanks in advance
> >> >> Cheers
> >> >>
> >> >> Il giorno gio 11 ago 2022 alle ore 16:45 Bill Miller III
> >> >> <brmilleriii.gmail.com> ha scritto:
> >> >> >
> >> >> > Honestly, I'm not sure what the snapshot-to-threads ratio should be. I'm also not sure if that's the problem. The issue actually might be the total frames and size of the system overwhelming the available RAM? The quasi-harmonic method requires loading all frames into memory (using cpptraj actually) all at once for the calculation, so that might actually be your issue more than snapshots per thread. Because I also don't think it makes use of cpptraj.MPI, so it probably just uses cpptraj in serial to load in all the frames, so there might be a memory availability issue depending on the size of your system.
> >> >> >
> >> >> > -Bill
> >> >> >
> >> >> > On Wed, Aug 10, 2022 at 10:24 AM Enrico Martinez <jmsstarlight.gmail.com> wrote:
> >> >> >>
> >> >> >> I gotcha, thank you so much!
> >> >> >> In my initial trajectory I have 3000 snapshots (corresponded to the
> >> >> >> 150 ns of production run).
> >> >> >> So I performed a quick test (using quasi-harmonic approximation of the
> >> >> >> entropy computing) taking 10 snapshots (interval=60) and 1 thread and
> >> >> >> it worked!. Then I tried mmpbsa.py.MPI with openMPI using 2 and 4
> >> >> >> threads and it seems that it works since the calculations passed to
> >> >> >> the step of the entropy computing (before, running the job using 90
> >> >> >> threads I had error just after finishing computing dH) .
> >> >> >> Finally I've just selected a smaller interval, increasing the number
> >> >> >> of snapshots to 50 and runned mmpbsa.py.MPI using only 5 threads and
> >> >> >> it's been working (while with 10, 16, 18 or 20 threads it produced
> >> >> >> error in the beginning of dS computing).
> >> >> >> How many snapshots / mpi threads should I use in my case ?
> >> >> >> Many thanks in advance
> >> >> >> Cheers
> >> >> >> Enrico
> >> >> >>
> >> >> >>
> >> >> >> Il giorno mer 10 ago 2022 alle ore 15:19 Bill Miller III
> >> >> >> <brmilleriii.gmail.com> ha scritto:
> >> >> >> >
> >> >> >> > You just have to be analyzing more frames than threads being used. Which it sounds like you’re doing. You’re interval said 10, but I didn’t know the total number of frames or total number of threads. Either way, it’s always a good idea to test in serial and parallel to know whether the error is possibly just a result of the parallelization or not.
> >> >> >> >
> >> >> >> > -Bill
> >> >> >> >
> >> >> >> >
> >> >> >> >
> >> >> >> > > On Aug 10, 2022, at 7:58 AM, Enrico Martinez <jmsstarlight.gmail.com> wrote:
> >> >> >> > >
> >> >> >> > > Dear all,
> >> >> >> > > Many thanks for your kind suggestions!
> >> >> >> > > Actually, for my computing I use a Dell workstation equipped with 96
> >> >> >> > > processors so for open-mpi I use 90 threads (=90 cpus). For instance,
> >> >> >> > > having a trajectory with 300 snapshots I divided it into 90 parts (per
> >> >> >> > > each processor) and then run mmpbsa using my script. Maybe I did not
> >> >> >> > > understand the parallelization well enough?? For example cpptraj using
> >> >> >> > > the same mpi strategy works quite well...
> >> >> >> > > Moreover , with MMPBSA.py.MPI I got dH computing (with 300 snapshots)
> >> >> >> > > just for 5 min...
> >> >> >> > > Cheers,
> >> >> >> > > Enrico
> >> >> >> > >
> >> >> >> > >> Il giorno mer 10 ago 2022 alle ore 14:14 Bill Miller III
> >> >> >> > >> <brmilleriii.gmail.com> ha scritto:
> >> >> >> > >>
> >> >> >> > >> Enrico,
> >> >> >> > >>
> >> >> >> > >> The error message says the error occurred on rank 68. Are you running this in parallel, and with how many threads? Rank 68 suggests you’re trying at least 69 threads, which is too many for only 10 frames (per your input file). You should use less (or equal to) threads than frames.
> >> >> >> > >>
> >> >> >> > >> I’d also suggest you try running the same calculation in serial and see if you get the same error.
> >> >> >> > >>
> >> >> >> > >> If I were debugging this problem, after doing the above, my next step would be to look at the _MMPBSA_avgcomplex.pdb file and compare it to your expected system. It says there is an atom mismatch of 8849 atoms when it expected 9333 atoms. That’s close to 500 atoms missing, so I’d suggest you try to figure out what those missing residues are and if there is something you can do to fix that.
> >> >> >> > >>
> >> >> >> > >> -Bill
> >> >> >> > >>
> >> >> >> > >>
> >> >> >> > >>
> >> >> >> > >>
> >> >> >> > >>
> >> >> >> > >>>> On Aug 10, 2022, at 4:19 AM, Enrico Martinez via AMBER <amber.ambermd.org> wrote:
> >> >> >> > >>>
> >> >> >> > >>> Dear Friends!
> >> >> >> > >>> Well, I still have problems with MMPBSA.py
> >> >> >> > >>> Firstly I created stripped topologies using
> >> >> >> > >>> ante-MMPBSA.py -p ${sim}/protein.prmtop -c ${sim}/protein.prmcrd -c
> >> >> >> > >>> complex.prmtop -r receptor.prmtop -l ligand.prmtop -s :WAT:Cl-:Na+:K+
> >> >> >> > >>> -n :lig > ante_${simulation}.log
> >> >> >> > >>>
> >> >> >> > >>> Then I checked all stripped topologies and trajectories, and
> >> >> >> > >>> everything was correct.
> >> >> >> > >>>
> >> >> >> > >>> Finally I run mmpbsa.py.MPI with this script:
> >> >> >> > >>> &general
> >> >> >> > >>> startframe= 1, interval=10, keep_files=2, entropy=1, netcdf=1
> >> >> >> > >>> /
> >> >> >> > >>> &gb
> >> >> >> > >>> igb=5, saltcon=0.150,
> >> >> >> > >>> /
> >> >> >> > >>> &decomp
> >> >> >> > >>> idecomp=1,
> >> >> >> > >>> dec_verbose=1,
> >> >> >> > >>> print_res="142,145,163,166,189"
> >> >> >> > >>> /
> >> >> >> > >>>
> >> >> >> > >>> 1) When I do calculations without entropy (entropy=0), everything goes OK!
> >> >> >> > >>>
> >> >> >> > >>>
> >> >> >> > >>> 2) when I add entropy=1 option to & general, it produces the following error:
> >> >> >> > >>>
> >> >> >> > >>> Error: PDB _MMPBSA_avgcomplex.pdb: No frames read. atom=8849 expected 9333.
> >> >> >> > >>> Error: Could not set up '_MMPBSA_avgcomplex.pdb' for reading.
> >> >> >> > >>> Error: DataSet_Coords_REF: Could not set up trajectory.
> >> >> >> > >>> 1 errors encountered reading input.
> >> >> >> > >>> Error: Error(s) occurred during execution.
> >> >> >> > >>> File "/home/gleb/amber22/bin/MMPBSA.py.MPI", line 100, in <module>
> >> >> >> > >>> app.run_mmpbsa()
> >> >> >> > >>> File "/home/gleb/amber22/lib/python3.9/site-packages/MMPBSA_mods/main.py",
> >> >> >> > >>> line 218, in run_mmpbsa
> >> >> >> > >>> self.calc_list.run(rank, self.stdout)
> >> >> >> > >>> File "/home/gleb/amber22/lib/python3.9/site-packages/MMPBSA_mods/calculation.py",
> >> >> >> > >>> line 82, in run
> >> >> >> > >>> calc.run(rank, stdout=stdout, stderr=stderr)
> >> >> >> > >>> File "/home/gleb/amber22/lib/python3.9/site-packages/MMPBSA_mods/calculation.py",
> >> >> >> > >>> line 375, in run
> >> >> >> > >>> Calculation.run(self, rank, stdout=self.output)
> >> >> >> > >>> File "/home/gleb/amber22/lib/python3.9/site-packages/MMPBSA_mods/calculation.py",
> >> >> >> > >>> line 156, in run
> >> >> >> > >>> raise CalcError('%s failed with prmtop %s!' % (self.program,
> >> >> >> > >>> CalcError: /home/gleb/amber22/bin/cpptraj failed with prmtop complex.prmtop!
> >> >> >> > >>> Error occurred on rank 68.
> >> >> >> > >>>
> >> >> >> > >>> Fatal Error!
> >> >> >> > >>> All files have been retained for your error investigation:
> >> >> >> > >>> You should begin by examining the output files of the first failed calculation.
> >> >> >> > >>> Consult the "Temporary Files" subsection of the MMPBSA.py chapter in the
> >> >> >> > >>> manual for file naming conventions.
> >> >> >> > >>> application called MPI_Abort(MPI_COMM_WORLD, 1) - process 68
> >> >> >> > >>>
> >> >> >> > >>> although I've mentioned the same script without entropy computing
> >> >> >> > >>> (using the same inputs) works perfectly...
> >> >> >> > >>>
> >> >> >> > >>> 3) In the case of entropy computing using NMA, I have another error
> >> >> >> > >>> mentioned earlier. How would it be possible to debug my system
> >> >> >> > >>> assuming that I've already double-checked all generated topologies and
> >> >> >> > >>> trajectories?
> >> >> >> > >>> Many thanks in advance!
> >> >> >> > >>> Enrico
> >> >> >> > >>>
> >> >> >> > >>>> Il giorno mar 9 ago 2022 alle ore 16:31 Enrico Martinez
> >> >> >> > >>>> <jmsstarlight.gmail.com> ha scritto:
> >> >> >> > >>>>
> >> >> >> > >>>> Right, thank you!
> >> >> >> > >>>> I am going to carry out some tests, I suppose the problem was related
> >> >> >> > >>>> to the re-imaging of the system. Otherwise, what options may be useful
> >> >> >> > >>>> for NMA section to overcome problems related to minimization?
> >> >> >> > >>>> Cheers,
> >> >> >> > >>>> Enrico
> >> >> >> > >>>>
> >> >> >> > >>>>> Il giorno mar 9 ago 2022 alle ore 16:17 Ray Luo <rluo.uci.edu> ha scritto:
> >> >> >> > >>>>>
> >> >> >> > >>>>> Enrico,
> >> >> >> > >>>>>
> >> >> >> > >>>>> "Line minimizer aborted" is related to your nmode option, so I don't
> >> >> >> > >>>>> think it's MPI related. You may want to test run a few snapshots with
> >> >> >> > >>>>> the serial mode to make sure the setup is correct before the MPI
> >> >> >> > >>>>> production run.
> >> >> >> > >>>>>
> >> >> >> > >>>>> Finally, please make sure your systems are correctly stripped before
> >> >> >> > >>>>> running again.
> >> >> >> > >>>>>
> >> >> >> > >>>>> Ray
> >> >> >> > >>>>> --
> >> >> >> > >>>>> Ray Luo, Ph.D.
> >> >> >> > >>>>> Professor of Structural Biology/Biochemistry/Biophysics,
> >> >> >> > >>>>> Chemical and Materials Physics, Chemical and Biomolecular Engineering,
> >> >> >> > >>>>> Biomedical Engineering, and Materials Science and Engineering
> >> >> >> > >>>>> Department of Molecular Biology and Biochemistry
> >> >> >> > >>>>> University of California, Irvine, CA 92697-3900
> >> >> >> > >>>>>
> >> >> >> > >>>>> On Tue, Aug 9, 2022 at 12:13 AM Enrico Martinez <jmsstarlight.gmail.com> wrote:
> >> >> >> > >>>>>>
> >> >> >> > >>>>>> Just finishing some tests I got the error on the step of the
> >> >> >> > >>>>>> calculation of dG for the complex:
> >> >> >> > >>>>>> Line minimizer aborted: step at upper bound 0.14208831
> >> >> >> > >>>>>> File "/home/gleb/amber22/bin/MMPBSA.py.MPI", line 100, in <module>
> >> >> >> > >>>>>> app.run_mmpbsa()
> >> >> >> > >>>>>> File "/home/gleb/amber22/lib/python3.9/site-packages/MMPBSA_mods/main.py",
> >> >> >> > >>>>>> line 218, in run_mmpbsa
> >> >> >> > >>>>>> self.calc_list.run(rank, self.stdout)
> >> >> >> > >>>>>> File "/home/gleb/amber22/lib/python3.9/site-packages/MMPBSA_mods/calculation.py",
> >> >> >> > >>>>>> line 82, in run
> >> >> >> > >>>>>> calc.run(rank, stdout=stdout, stderr=stderr)
> >> >> >> > >>>>>> File "/home/gleb/amber22/lib/python3.9/site-packages/MMPBSA_mods/calculation.py",
> >> >> >> > >>>>>> line 321, in run
> >> >> >> > >>>>>> Calculation.run(self, rank, stdout=self.output % rank)
> >> >> >> > >>>>>> File "/home/gleb/amber22/lib/python3.9/site-packages/MMPBSA_mods/calculation.py",
> >> >> >> > >>>>>> line 156, in run
> >> >> >> > >>>>>> raise CalcError('%s failed with prmtop %s!' % (self.program,
> >> >> >> > >>>>>> CalcError: /home/gleb/amber22/bin/mmpbsa_py_nabnmode failed with
> >> >> >> > >>>>>> prmtop complex.prmtop!
> >> >> >> > >>>>>> Error occurred on rank 46.
> >> >> >> > >>>>>>
> >> >> >> > >>>>>> Fatal Error!
> >> >> >> > >>>>>> All files have been retained for your error investigation:
> >> >> >> > >>>>>> You should begin by examining the output files of the first failed calculation.
> >> >> >> > >>>>>> Consult the "Temporary Files" subsection of the MMPBSA.py chapter in the
> >> >> >> > >>>>>> manual for file naming conventions.
> >> >> >> > >>>>>> application called MPI_Abort(MPI_COMM_WORLD, 1) - process 46
> >> >> >> > >>>>>>
> >> >> >> > >>>>>> may it be related to the open mpi problem or rather something wrong
> >> >> >> > >>>>>> with the system ( I run it on the system that had not been
> >> >> >> > >>>>>> reimaged..)?
> >> >> >> > >>>>>>
> >> >> >> > >>>>>> Il giorno lun 8 ago 2022 alle ore 19:20 Ray Luo <rluo.uci.edu> ha scritto:
> >> >> >> > >>>>>>>
> >> >> >> > >>>>>>> Hi Enrico,
> >> >> >> > >>>>>>>
> >> >> >> > >>>>>>> Here are my suggestions regarding your questions ...
> >> >> >> > >>>>>>>
> >> >> >> > >>>>>>>> 1) Dealing with the modeling of the dimeric protein where the ligand
> >> >> >> > >>>>>>>> is bound to one of the monomers, do I need to re-image my system prior
> >> >> >> > >>>>>>>> to the post-processing (avoiding jumping parts of the protein across
> >> >> >> > >>>>>>>> the PB) ?
> >> >> >> > >>>>>>>
> >> >> >> > >>>>>>> Yes you'll need to make sure all proteins are in the primary cell
> >> >> >> > >>>>>>> before stripping waters/ions etc. To make sure the system looks right,
> >> >> >> > >>>>>>> I strongly recommend visualizing the stripped trajectories before
> >> >> >> > >>>>>>> running the job.
> >> >> >> > >>>>>>>
> >> >> >> > >>>>>>>> 2) What snapshot density is required for the calculations? For
> >> >> >> > >>>>>>>> example, dealing with the 150ns trajectory where each snapshot was
> >> >> >> > >>>>>>>> saved every 25000 steps, I have around 6000 snapshots. In order to
> >> >> >> > >>>>>>>> speed up the analysis, would it be OK to take each 5th snapshot for
> >> >> >> > >>>>>>>> the calculations of dH and each 10th for the dS ?
> >> >> >> > >>>>>>>
> >> >> >> > >>>>>>> In general, you can use fewer snapshots to save time. The con is that
> >> >> >> > >>>>>>> you'd get higher uncertainties, i.e. standard errors.
> >> >> >> > >>>>>>>
> >> >> >> > >>>>>>> All the best,
> >> >> >> > >>>>>>> Ray
> >> >> >> > >>>>>>> --
> >> >> >> > >>>>>>> Ray Luo, Ph.D.
> >> >> >> > >>>>>>> Professor of Structural Biology/Biochemistry/Biophysics,
> >> >> >> > >>>>>>> Chemical and Materials Physics, Chemical and Biomolecular Engineering,
> >> >> >> > >>>>>>> Biomedical Engineering, and Materials Science and Engineering
> >> >> >> > >>>>>>> Department of Molecular Biology and Biochemistry
> >> >> >> > >>>>>>> University of California, Irvine, CA 92697-3900
> >> >> >> > >>>>>>>
> >> >> >> > >>>>>>> On Mon, Aug 8, 2022 at 7:49 AM Enrico Martinez via AMBER
> >> >> >> > >>>>>>> <amber.ambermd.org> wrote:
> >> >> >> > >>>>>>>>
> >> >> >> > >>>>>>>> Here is my script
> >> >> >> > >>>>>>>> &general
> >> >> >> > >>>>>>>> startframe= 1, interval=${int}, keep_files=2,
> >> >> >> > >>>>>>>> strip_mask=:WAT:Cl-:Na+:K+, netcdf=1
> >> >> >> > >>>>>>>> /
> >> >> >> > >>>>>>>> &gb
> >> >> >> > >>>>>>>> igb=5, saltcon=0.150,
> >> >> >> > >>>>>>>> /
> >> >> >> > >>>>>>>> &nmode
> >> >> >> > >>>>>>>> nmstartframe=1, nminterval=5, nmode_igb=1, drms=0.05, nmode_istrng=0.15,
> >> >> >> > >>>>>>>> /
> >> >> >> > >>>>>>>> &decomp
> >> >> >> > >>>>>>>> idecomp=1,
> >> >> >> > >>>>>>>> dec_verbose=1,
> >> >> >> > >>>>>>>> /
> >> >> >> > >>>>>>>>
> >> >> >> > >>>>>>>> Many thanks in advance!
> >> >> >> > >>>>>>>> Cheers
> >> >> >> > >>>>>>>> Enrico
> >> >> >> > >>>>>>>>
> >> >> >> > >>>>>>>> _______________________________________________
> >> >> >> > >>>>>>>> AMBER mailing list
> >> >> >> > >>>>>>>> AMBER.ambermd.org
> >> >> >> > >>>>>>>> https://urldefense.com/v3/__http://lists.ambermd.org/mailman/listinfo/amber__;!!CzAuKJ42GuquVTTmVmPViYEvSg!MozQWh4unot4uyigV3FICNj-0nADRKcRD8e-wAuplUDGC-vuu938BB2NqCwqY1uSi8vPhnX1fj4$
> >> >> >> > >>>
> >> >> >> > >>> _______________________________________________
> >> >> >> > >>> AMBER mailing list
> >> >> >> > >>> AMBER.ambermd.org
> >> >> >> > >>> http://lists.ambermd.org/mailman/listinfo/amber
> >> >> >
> >> >> >
> >> >> >
> >> >> > --
> >> >> > Bill Miller III
> >> >> > Assistant Professor of Chemistry
> >> >> > Truman State University
> >> >> > 417-549-0952
> >> >
> >> >
> >> >
> >> > --
> >> > Bill Miller III
> >> > Assistant Professor of Chemistry
> >> > Truman State University
> >> > 417-549-0952
> >
> >
> >
> > --
> > Bill Miller III
> > Assistant Professor of Chemistry
> > Truman State University
> > 417-549-0952
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Aug 18 2022 - 01:00:04 PDT
