Previous Thread
Next Thread
Print Thread
Page 1 of 3 1 2 3
Joined: Oct 2011
Posts: 50
M
Forum Member
OP Offline
Forum Member
M
Joined: Oct 2011
Posts: 50
Dear Sir,
I am using GPU-OpenMM interaface for running dynamics in GPU. Protein is out of box in properly equilibrated solvated box when i performing MD run . Please suggest me why protein is coming out of box only when i use openmm . in normal MD there is no centering problem occured.

Below i am writing MD code whic i used in Openmm-gpu run.

!Image centering by residue
IMAGE BYSEG XCEN @xcen YCEN @ycen ZCEN @zcen select (segid PROA .or. segid PROB .or. segid HETA) end
IMAGE BYRESID XCEN @xcen YCEN @ycen ZCEN @zcen sele resname TIP3 end
IMAGE BYRESID XCEN @xcen YCEN @ycen ZCEN @zcen sele ( segid @posid .or. segid @negid ) end
! Nonbonded Options
set nsteps = 1000
set cutoff = 11
set ctofnb = 8
set ctonnb = 7.5
set kappa = 0.3308 ! Consistent with cutofnb and fftx,y,z
calc cutim = @cutoff
!nbonds atom vatom vfswitch bycb -
! ctonnb 10.0 ctofnb 12.0 cutnb 16.0 cutim 16.0 -
! inbfrq -1 imgfrq -1 wmin 1.0 cdie eps 1.0 -
! ewald pmew fftx @fftx ffty @ffty fftz @fftz kappa .34 spline order 6
faster on
energy eps 1.0 cutnb @cutoff cutim @cutim -
ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx 64 ffty 64 fftz 64

shake fast bonh tol 1.0e-8 para
!use a restraint to place center of mass of the molecules near the origin
MMFP
GEO rcm sphere -
Xref @xcen Yref @ycen Zref @zcen XDIR 1.0 YDIR 1.0 ZDIR 1.0 -
harmonic FORCE 10.0 select .not. ( hydrogen .or. resname TIP3 .or. segid @posid .or. segid @negid ) end
END
cons harm force 8 exponent 3 select hydrogen end
coor translate xdir 1.3 select all end
set nsteps = 1000
open write unit 12 card name step5.1_production.rst
open write unit 13 file name step5.1_production.dcd
set echeck = echeck -1
dynamics leap start timestep 0.001 -
nstep @nsteps nprint 100 iprfrq 1000 -
firstt 300 finalt 300 twindl -5.0 twindh 5.0 -
ichecw 0 teminc 30.0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck -
iunread -1 iunwri 12 iuncrd 13 -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa .34 pme order 6 fftx @fftx ffty @ffty fftz @fftz ntrfq @nsteps -
omm langevin gamma 5.0 -
prmc pref 1 iprsfrq 25


Thanks & Regards
Manish Kesherwani

Joined: Sep 2003
Posts: 8,606
Likes: 24
rmv Online Content
Forum Member
Online Content
Forum Member
Joined: Sep 2003
Posts: 8,606
Likes: 24
It may depend on the values used for @xcen, etc. for the MMFP restraint; I suggest trying w/o restraints and then find the protein center. I found that OpenMM in CHARMM seems to shift the Cartesian coordinates such that the minimum corner of the unit cell is at the origin, instead of the origin being the center. Also, since the OpenMM interface is still somewhat developmental, it's not clear if all of the restraints in CHARMM are fully supported; the documentation does not address the issue.


Rick Venable
computational chemist

Joined: Oct 2011
Posts: 50
M
Forum Member
OP Offline
Forum Member
M
Joined: Oct 2011
Posts: 50
Thanks sir for suggestion. i checked without MMFP restraint on protein Results are same protein is out of solvated box.
How to solve this problem please give suggestion

I also xcen=0 is giving in image centering


output is

DCNTRL> Langevin integration requested.
DCNTRL> OpenMM interface requested for energy and dynamics calculations.
DCNTRL> Constant temperature w/ OpenMM using Langevin heatbath requested.
DCNTRL> Reference heatbath temperature is 300.000K
DCNTRL> Using friction coefficient value of 10.000 ps^-1
DCNTRL> CPT dynamics through OpenMM interface requested using Langevin heatbath.
DCNTRL> MC barostat coupled to reference pressure 1.00 atmospheres.
DCNTRL> MC barostat using reference temperature 300.00 K.
DCNTRL> MC barostat volume move attempted every 25 timesteps.
NSTEP = 1000 NSAVC = 10 NSAVV = 10
ISCALE = 0 ISCVEL = 0 IASORS = 1
IASVEL = 1 ICHECW = 0 NTRFRQ = 1000
IHTFRQ = 0 IEQFRQ = 0 NPRINT = 100
INBFRQ = -1 IHBFRQ = 0 IPRFRQ = 1000
ILBFRQ = 0 IMGFRQ = -1
ISVFRQ = 0 NCYCLE = 1000 NSNOS = 10
FIRSTT = 300.000 TEMINC = 30.000 TSTRUC = -999.000
FINALT = 300.000 TWINDH = 5.000 TWINDL = -5.000

TIME STEP = 4.09097E-02 AKMA 2.00000E-03 PS

RANDOM NUM. GEN. SEED(S) = 1151852549 1151852549 1151852549 1151852549

SHAKE TOLERANCE = 0.10000E-04
NUMBER OF DEGREES OF FREEDOM = 61928

GAUSSIAN OPTION IS 1
VELOCITIES ASSIGNED AT TEMPERATURE = 375.0000
SEED FOR RANDOM NUMBER GENERATOR IS:
SEEDS> 1151852549 1151852549 1151852549 1151852549

DETAILS ABOUT CENTRE OF MASS
POSITION : -0.11498689 1.58487956E-02 -0.12708586
VELOCITY : -2.55257175E-03 2.51749404E-03 -1.34131645E-03
ANGULAR MOMENTUM : 3326.1685 3772.3654 9957.1191
KINETIC ENERGY : 1.3681834
Initializing OpenMM context
OpenMM version 5.0
Specifying OpenMM Ewald error tolerance 1.911E-04
OpenMM will use KAPPA of 3.117E-01 A^(-1)
OpenMM nodes in PME mesh will be closest
product of 2, 3 and 5 to 80 80 80
Using OpenMM platform OpenCL
OpenCLPlatformIndex = 0
OpenCLPrecision = single
OpenCLDeviceIndex = 0



Thanks & Regards
Manish Kesherwani

Last edited by Manish Kesherwan; 07/02/13 04:22 PM.
Joined: Sep 2003
Posts: 8,606
Likes: 24
rmv Online Content
Forum Member
Online Content
Forum Member
Joined: Sep 2003
Posts: 8,606
Likes: 24
Sorry, but the output you've posted doesn't tell me anything new about the problem, and doesn't confirm the lack of restraints. I don't have a lot of OpenMM experience myself, and have not observed the problem you seem to be having.

Besides the modified cutoffs and reduced SHAKE tolerance (usually bad ideas), I noticed these curious commands in the original input--

cons harm force 8 exponent 3 select hydrogen end
coor translate xdir 1.3 select all end


These are very odd and puzzling.


Rick Venable
computational chemist

Joined: Oct 2011
Posts: 50
M
Forum Member
OP Offline
Forum Member
M
Joined: Oct 2011
Posts: 50
sorry sir i deleted these two commands after that also there is no changes problems are remained. I also didn't get these problem in normal CPU run. it may be box decomposition in memory allocation for GPU and unable to assign interaction between segment of system.
I have C37b2 version of charmm.

Please can you suggest some charmm-openmm interface developer or some sample script.

Thanks & Regards
Manish Kesherwani

Joined: Sep 2003
Posts: 8,606
Likes: 24
rmv Online Content
Forum Member
Online Content
Forum Member
Joined: Sep 2003
Posts: 8,606
Likes: 24
I'm afraid I don't know much more than what is contained in openmm.doc


Rick Venable
computational chemist

Joined: Sep 2003
Posts: 4,850
Likes: 10
Forum Member
Online Content
Forum Member
Joined: Sep 2003
Posts: 4,850
Likes: 10
And what is the error you get?
Posting actual input and output (brief and relevant) is highly recommended.


Lennart Nilsson
Karolinska Institutet
Stockholm, Sweden
Joined: Oct 2011
Posts: 50
M
Forum Member
OP Offline
Forum Member
M
Joined: Oct 2011
Posts: 50
Thanks for repling sir

I used this input code for run dynamics and three attached file first input equilibrated structure file second output pdb and third log file

Please suggest me what is wrong in input file for run in openmm -GPU code.

Input file:


CRYSTAL DEFINE @XTLtype @A @B @C @alpha @beta @gamma
CRYSTAL READ UNIT 10 CARD

!Image centering by residue
IMAGE BYSEG XCEN 0 YCEN 0 ZCEN 0 select (segid PROA .or. segid PROB .or. segid HETA) end
IMAGE BYRESID XCEN 0 YCEN 0 ZCEN 0 sele resname TIP3 end
IMAGE BYRESID XCEN 0 YCEN 0 ZCEN 0 sele ( segid @posid .or. segid @negid ) end
set cutoff = 11
set ctofnb = 9
set ctonnb = 9
calc cutim = @cutoff
set kappa = 0.32

faster on
nbonds eps 1.0 cutnb @cutoff cutim @cutim -
ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx @fftx ffty @ffty fftz @fftz

shake fast bonh tol 1.0e-5 para

set echeck = echeck -1
set nsteps = 1000
open write unit 12 card name step5.1_production.rst
open write unit 13 file name step5.1_production.dcd
! Run dynamics in periodic box
dynamics leap start timestep 0.002 -
nstep @nsteps nprint 100 iprfrq 1000 -
firstt 300 finalt 300 twindl -5.0 twindh 5.0 -
ichecw 0 teminc 30.0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck -
iunread -1 iunwri 12 iuncrd 13 -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx @fftx ffty @ffty fftz @fftz ntrfq @nsteps - !PME
omm langevin gamma 10 prmc iprssfrq 25 pref 1


Thanks & Regards
Manish Kesherwani

Attached Images
charmm-gui.tar.gz (952.6 KB, 881 downloads)
Joined: Sep 2003
Posts: 4,850
Likes: 10
Forum Member
Online Content
Forum Member
Joined: Sep 2003
Posts: 4,850
Likes: 10
Well, I don't know what to say. You are apparently trying to run before you can crawl. There is a specific error message in your output file, which is unrelated to GPU-usage and water centering, as far as I can tell. Since you use CHARMM-GUI, I think your best option is to ask for support from the CHARMM-GUI people (I am not familiar with the peculiarities of CHARMM-GUI).


Lennart Nilsson
Karolinska Institutet
Stockholm, Sweden
Joined: Oct 2011
Posts: 50
M
Forum Member
OP Offline
Forum Member
M
Joined: Oct 2011
Posts: 50
Sorry sir i corrected that error in command now it's script are running in GPU but still output structure is not centering.

sir i modified totally script according to run in GPU. i removed MMFF contraint and nonbonded parametr from script. so it's changed from charmmgui given script. i don't know how to center the molecule.

Please give some suggestions

Here is input command and attached output file.

dynamics leap start timestep 0.002 -
nstep @nsteps nprint 100 iprfrq 1000 -
firstt 300 finalt 300 twindl -5.0 twindh 5.0 -
ichecw 0 teminc 30.0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 -
ilbfrq 0 inbfrq -1 imgfrq -1 @echeck -
iunread -1 iunwri 12 iuncrd 13 -
eps 1.0 cutnb @cutoff cutim @cutim ctofnb @ctofnb ctonnb @ctonnb vswi -
ewald kappa @kappa pme order 4 fftx @fftx ffty @ffty fftz @fftz ntrfq @nsteps - !PME
omm langevin gamma 10 prmc iprssfrq 25 pref 1

Attached Images
step5.txt (61.4 KB, 1053 downloads)
Joined: Sep 2003
Posts: 8,606
Likes: 24
rmv Online Content
Forum Member
Online Content
Forum Member
Joined: Sep 2003
Posts: 8,606
Likes: 24
In the output, the IMAGE BYSEG command is commented out, which may be one reason the protein is not placed properly.

All of the extra stuff from CHARMM-GUI tends to complicate problem diagnosis, esp. the vast array of RTF and PARAM files being read.


Rick Venable
computational chemist

Joined: Oct 2011
Posts: 50
M
Forum Member
OP Offline
Forum Member
M
Joined: Oct 2011
Posts: 50
I also used IMAGEing bysegment but it's not showing any centering of protein molecule in water box.

I don't know why there is no contraint on protein for centering in openmm.

Thanks & Regards
Manish Kesherwani

Attached Images
log.txt (63.14 KB, 953 downloads)
step5.txt (61.4 KB, 1422 downloads)
Joined: Sep 2003
Posts: 4,850
Likes: 10
Forum Member
Online Content
Forum Member
Joined: Sep 2003
Posts: 4,850
Likes: 10
I suggest you ask the CHARMM-GUI support - they should know how their script is supposed to work.


Lennart Nilsson
Karolinska Institutet
Stockholm, Sweden
Joined: Sep 2003
Posts: 8,606
Likes: 24
rmv Online Content
Forum Member
Online Content
Forum Member
Joined: Sep 2003
Posts: 8,606
Likes: 24
It is possible the script from CHARMM-GUI has not been tested with OpenMM, as the latter is a fairly recent development which requires custom hardware.

Consider setting up a much simpler test case yourself, with just a short peptide in a cube of water, using only the protein parameters with the water_ions stream file.


Rick Venable
computational chemist

Joined: Oct 2011
Posts: 50
M
Forum Member
OP Offline
Forum Member
M
Joined: Oct 2011
Posts: 50
Thanks Rick sir i will try with simple peptide with solvent box.

Thanks & Regards
Manish Kesherwani

Joined: Oct 2011
Posts: 50
M
Forum Member
OP Offline
Forum Member
M
Joined: Oct 2011
Posts: 50
Thanks sir it's working now with imaging of protein segment during run of dynamics.

Thanks & Regards
Manish Kesherwani

Joined: Sep 2006
Posts: 146
S
Forum Member
Offline
Forum Member
S
Joined: Sep 2006
Posts: 146
Is MMFP actually supported with OpenMM?

Joined: Oct 2011
Posts: 50
M
Forum Member
OP Offline
Forum Member
M
Joined: Oct 2011
Posts: 50
Sir i tried with energy command along with omm then it's giving mmff contribution of energy . but i faced one problem that protein molecule is not coming in center of box during simulation but afer generating trajactory we can align frames with protein molecule then molecule will be in centre. If you have some other solution regarding centering of protein molecule in waterbox please sugggest me. Did you tried simulation with openmm. how was your trajactory.

Joined: Sep 2003
Posts: 8,606
Likes: 24
rmv Online Content
Forum Member
Online Content
Forum Member
Joined: Sep 2003
Posts: 8,606
Likes: 24
As I noted above, in my initial reply--

"It may depend on the values used for @xcen, etc. for the MMFP restraint; I suggest trying w/o restraints and then find the protein center. I found that OpenMM in CHARMM seems to shift the Cartesian coordinates such that the minimum corner of the unit cell is at the origin, instead of the origin being the center. Also, since the OpenMM interface is still somewhat developmental, it's not clear if all of the restraints in CHARMM are fully supported; the documentation does not address the issue."


Rick Venable
computational chemist

Joined: Oct 2011
Posts: 50
M
Forum Member
OP Offline
Forum Member
M
Joined: Oct 2011
Posts: 50
Good Day Sir,

I am running equilibration dynamics in membrane protein using NVIDIA TESLA GPU card(openmm -charmm). when i am using multigpu(more than one gpu) then energy drift is too high around( 1000000 Kcal). Can u specify how we can rectify this.

Please give some suggextion

Joined: Oct 2011
Posts: 50
M
Forum Member
OP Offline
Forum Member
M
Joined: Oct 2011
Posts: 50
I used this command for equilibration in NPT using openmm

dyna leap -
restart nstep 1000000 timestp 0.001 -
inbfrq -1 imgfrq -1 ihbfrq 0 ilbfrq 0 -
iunread 11 iunwrite 12 iuncrd 13 -
nsavc 1000 nsavvelo 0 ntrfq 1000 -
nprint 1000 iprfrq 1000 -
firstt @temp finalt @temp twindl -5.0 twindh 5.0 -
ichecw 0 teminc 30.0 ihtfrq 0 ieqfrq 0 -
iasors 1 iasvel 1 iscvel 0 @echeck bycb -
eps 1.0 ctonnb 10.0 ctofnb 12.0 cutnb 16.0 cutim 16.0 vswi -
ewald pmew fftx @fftx ffty @ffty fftz @fftz kappa .34 spline order 6 -
omm langevin gamma 10 - ! turn on openmm, set-up Langevin
prmc pref 1 iprsfrq 25 ! set-up MC barostat at 1 atm, move attempt / 25 steps


Normal CPU run

DYNA CPT leap restart nstep @nstep timestep 0.001 echeck 9999. -
nprint 100 iprfrq 5000 ntrfrq 5000 -
iunrea 11 iunwri 12 iuncrd 13 iunvel -1 kunit -1 -
nsavc 100 nsavv 0 -
PCONS pint pref 1.0 pmass @Pmass pgamma 20.0 -
HOOVER reft @temp tmass 2000.0 tbath @temp firstt @temp

Please help me in running parallel GPU

Joined: Sep 2003
Posts: 8,606
Likes: 24
rmv Online Content
Forum Member
Online Content
Forum Member
Joined: Sep 2003
Posts: 8,606
Likes: 24
Since the questions and answers in this post mostly involve the OpenMM interface, it was moved here to this forum.

The energy drift is a function of the numerical precision, and has been shown to be fairly bad when using a GPU with single precision. Mixed precision can improve that somewhat, and double precision is better, but costs more, in terms of time to completion. The env var OPENMM_PRECISION can be used to control the precision.

The default prmc code does not implement an anisotropic pressure tensor, is therefore not appropriate for simulating membrane bilayer based systems. Code has been under development for this, but it does not appear to be in any of the public beta release versions of OpenMM/CHARMM.


Rick Venable
computational chemist

Joined: Oct 2011
Posts: 50
M
Forum Member
OP Offline
Forum Member
M
Joined: Oct 2011
Posts: 50
Thanks for your reply about openmm prmc option for simulating membrane system.
If anisotropic pressure tensor is not supporting then what method can be applied for simulating membrane system.

I am using charmm 37b2.

Please suggest me.

Joined: Sep 2003
Posts: 8,606
Likes: 24
rmv Online Content
Forum Member
Online Content
Forum Member
Joined: Sep 2003
Posts: 8,606
Likes: 24
The DOMDEC code in CHARMM (domdec.doc) provides the best combination of speed and accuracy, but what's best for your case may depend both on available hardware and the molecular system.


Rick Venable
computational chemist

Joined: Jul 2019
Posts: 27
C
Forum Member
Offline
Forum Member
C
Joined: Jul 2019
Posts: 27
Originally Posted By: slaw
Is MMFP actually supported with OpenMM?


Using the latest version of OpenMM, it doesn't appear to be:

Quote:
***** LEVEL -1 WARNING FROM *****
***** Number of atoms in restraint too large/geometric shape not supported


Where the MMFP block generated by CHARMM-GUI applying a restraint to non-waters looks like this:

Code:
MMFP
GEO rcm sphere -
    Xref @xcen Yref @ycen Zref @zcen XDIR 1.0 YDIR 1.0 ZDIR 1.0 -
    harmonic FORCE 1.0 select .not. ( hydrogen .or. resname TIP3 .or. segid @posid .or. segid @negid ) end
END


My system is a small peptide system (12 AA's) so I doubt size is the issue. When run on CPUs, the peptide is nicely centred in the water box. Commenting out the block means the simulation runs using OpenMM acceleration but the peptide flaps around outside the centred water box.

I note in a more recent post ( CHARMM Forum | How to maintain plane orientation), the suggestion of CONS HARM ABSO as the solution.

So, if I wanted to apply a harmonic constraint as per the manual CONS to mimic effect of the MMFP block, perhaps something like this would do?

Quote:
CONS HARM ABSO 1.0 FORCE 1.0 SELE .not. ( hydrogen .or. resname TIP3 .or. segid @posid .or. segid @negid ) END COMP

Last edited by Corey Taylor; 08/08/19 09:12 AM.
Joined: Sep 2003
Posts: 8,606
Likes: 24
rmv Online Content
Forum Member
Online Content
Forum Member
Joined: Sep 2003
Posts: 8,606
Likes: 24
An issue with that is the suppression of conformational changes; you could just restrain a single atom, e.g. a Calpha near the middle of the sequence.

(I should add that my preference has been to use the Python interface to OpenMM, which is somewhat notable as I've been avoiding Python for years.)


Rick Venable
computational chemist

Joined: Jul 2019
Posts: 27
C
Forum Member
Offline
Forum Member
C
Joined: Jul 2019
Posts: 27
Cheers again for the tip - good point re: conformational changes.

And do you mean this (A MD/GCMC OpenMM/XML interface) with respect to the interface?

Joined: Sep 2003
Posts: 8,606
Likes: 24
rmv Online Content
Forum Member
Online Content
Forum Member
Joined: Sep 2003
Posts: 8,606
Likes: 24
I'm aware of a couple different python approaches: one using openmmtools, and another from CHARMM-GUI. This seems to be yet another approach, and I'll wager there may be others.


Rick Venable
computational chemist

Page 1 of 3 1 2 3

Moderated by  lennart, rmv 

Link Copied to Clipboard
Powered by UBB.threads™ PHP Forum Software 7.7.5
(Release build 20201027)
Responsive Width:

PHP: 7.3.31-1~deb10u1 Page Time: 0.020s Queries: 73 (0.012s) Memory: 0.8929 MB (Peak: 1.0767 MB) Data Comp: Off Server Time: 2022-10-06 17:42:58 UTC
Valid HTML 5 and Valid CSS