Hi Juraj,
I have been trying to calculate the absolute binding free energies of a small molecule to my kinase protein and I have faced issues too with the NPT equilibration steps on GPUs. The cpu version runs stable. Does that indicate there may be a bug in the GPU code using NPT ensemble?
Well since I could not equilibrate my complex at higher lambdas, I am stuck at the phase and I m trying to make things work.
I have a paper which does TI on gPUs and succeeds; thought it might help you in setting up the protocol better.
This is the link
https://doi.org/10.1021/acs.jcim.9b00105
Thanks,
Debarati
From: Juraj Dobias<mailto:juraj.dobias.uochb.cas.cz>
Sent: 10 April 2020 02:04
To: amber.ambermd.org<mailto:amber.ambermd.org>
Subject: Re: [AMBER] Ligand leaving binding site at high lambdas using TI softcore
Hi Callum,
Thanks for your response.
I can confirm, that when ligand change is smaller system is more stable.
For example changing Cl on benzene to H was stable until lambda 0.7.
I am also using Amber18 GPU. I use CPU code for NVT heating and first
NPT equilibration, as GPU code had problems with large cell volume changes.
Sometimes ligands became entangled inside protein and then rip it apart.
I think decoupled ligand goes deeper to the protein which would be
normal, but then coupled ligand coordinates are resynchronized with
decoupled one, protein is deformed, allowing decoupled ligand to go even
deeper and process repeats. At low lambdas this is not happening because
coordinates are synchronized always to first ligand which is in case of
low lambda more coupled one.
After your suggestion, I tried 100 ns NVT production run and ligand was
stable at lambda 0.9 which is great. Then I also tried same NPT
production run as before but using CPU pmemd.MPI instead of pmemd.cuda
and again ligand was stable. This might suggest a bug in GPU code using
NPT ensemble and TI.
Juraj
On 09. 04. 20 21:00, Dickson, Callum wrote:
> Hi Juraj,
>
> I have recently been trying out TI simulations and have observed the same behaviour. I am by no means an expert at free energy runs so I'd be interested to hear advice from others.
>
> This was occurring for me when the TI regions were large (different transformations with smaller TI regions did not result in ligands flying away). In NPT ensemble, certain ligands would immediately leave the pocket. The same system in NVT ensemble was more stable, without the ligand drifting away. I am using the Amber18 GPU code, fully patched.
>
> Do you see this behaviour also with NVT production run?
>
> Best,
> Callum
>
>
> -----Original Message-----
> From: Juraj Dobias <juraj.dobias.uochb.cas.cz>
> Sent: Thursday, April 9, 2020 1:30 PM
> To: amber.ambermd.org
> Subject: Re: [AMBER] Ligand leaving binding site at high lambdas using TI softcore
>
> To make it clearer. As I understand TI, simulation at 0.1 lambda should be in principle equivalent to simulation at 0.9 lambda with switched endstates, is it correct?
>
> I think problem is that coordinates of common atoms in each TI regions are changed independently during simulation (I thought it is not
> possible) and then resynchronized, so that coordinates from T1 are copied to T2.
>
> Coordinates for common atoms in starting pdb are exactly equal, but I observe these warnings starting in equil3 (restraining for is smaller than before)
>
> WARNING: Local coordinate 6005 differs from partner coordinate 6035 !
> Deviation is small, changing partner coordinate.
>
> If I define whole ligands as softcore, simulation works as expected. But it is not ideal, because low coupled ligand becomes unstable.
>
>
> On 08. 04. 20 22:39, Juraj Dobias wrote:
>> Dear all,
>>
>> I am trying to utilize TI in amber in our medicinal chemistry project.
>> I successfully did tutorial of transforming benzene to phenol, but I
>> am struggling for a week with my system.
>>
>> Problem is that ligands have higher tendency to immediately leave
>> binding site after releasing restraints at higher lambda values.
>>
>> My workflow is:
>>
>> 10000 steps of minimization (https://urldefense.proofpoint.com/v2/url?u=http-3A__min.in&d=DwIGaQ&c=ZbgFmJjg4pdtrnL2HUJUDw&r=HwsFjSfOtLupDR-NuCP430rdz1DD2LkJxNM3BsKSjrw&m=ENVyFj6_oD0p_qmj-WymfucA1YgEb6nnnRKPbWF9MfA&s=e0q7Xe4kHAxgfod1sPVqnh1HKrl2UP7fk2UxfyKkVCE&e= )
>>
>> 50 ps heating with restrained solute heavy atoms with force constant
>> 100 (https://urldefense.proofpoint.com/v2/url?u=http-3A__heat.in&d=DwIGaQ&c=ZbgFmJjg4pdtrnL2HUJUDw&r=HwsFjSfOtLupDR-NuCP430rdz1DD2LkJxNM3BsKSjrw&m=ENVyFj6_oD0p_qmj-WymfucA1YgEb6nnnRKPbWF9MfA&s=NLeEZSyD66QJIdPVE5AbQelZVRFwVKDwhLHer2UntN0&e= )
>>
>> 50 ps npt with restrained solute heavy atoms with force constant 100
>> (https://urldefense.proofpoint.com/v2/url?u=http-3A__equil1.in&d=DwIGaQ&c=ZbgFmJjg4pdtrnL2HUJUDw&r=HwsFjSfOtLupDR-NuCP430rdz1DD2LkJxNM3BsKSjrw&m=ENVyFj6_oD0p_qmj-WymfucA1YgEb6nnnRKPbWF9MfA&s=hPB98VnDPjWWhLtRU6TZvLwlkRt0tuHJPlLztaiC5yM&e= )
>>
>> 100 ps npt with restrained solute heavy atoms with force constant 10
>> (https://urldefense.proofpoint.com/v2/url?u=http-3A__equil2.in&d=DwIGaQ&c=ZbgFmJjg4pdtrnL2HUJUDw&r=HwsFjSfOtLupDR-NuCP430rdz1DD2LkJxNM3BsKSjrw&m=ENVyFj6_oD0p_qmj-WymfucA1YgEb6nnnRKPbWF9MfA&s=xGceI4OdNjaGchBjvtNilphqBEDaMp3B2HpDUHg6IhM&e= )
>>
>> 100 ps npt with restrained protein backbone and core TI region with
>> force constant 10 (https://urldefense.proofpoint.com/v2/url?u=http-3A__equil3.in&d=DwIGaQ&c=ZbgFmJjg4pdtrnL2HUJUDw&r=HwsFjSfOtLupDR-NuCP430rdz1DD2LkJxNM3BsKSjrw&m=ENVyFj6_oD0p_qmj-WymfucA1YgEb6nnnRKPbWF9MfA&s=BFfEgRfukbu-Hu1ZkCjo5KJi4OvfqxT9eiXg0ZcchWI&e= )
>>
>> 100 ps npt with restrained protein backbone and core TI region with
>> force constant 1 (https://urldefense.proofpoint.com/v2/url?u=http-3A__equil4.in&d=DwIGaQ&c=ZbgFmJjg4pdtrnL2HUJUDw&r=HwsFjSfOtLupDR-NuCP430rdz1DD2LkJxNM3BsKSjrw&m=ENVyFj6_oD0p_qmj-WymfucA1YgEb6nnnRKPbWF9MfA&s=3KwNrn-fMGnA_Fj1GdlUYqjNEzAo4ytUDD_cdecr6GM&e= )
>>
>> 100 ps npt without restraints (https://urldefense.proofpoint.com/v2/url?u=http-3A__prod.in&d=DwIGaQ&c=ZbgFmJjg4pdtrnL2HUJUDw&r=HwsFjSfOtLupDR-NuCP430rdz1DD2LkJxNM3BsKSjrw&m=ENVyFj6_oD0p_qmj-WymfucA1YgEb6nnnRKPbWF9MfA&s=7XD5Q06IvkvqIxudn9B0QZ1dIyUXTrF-R0uCizZR9bI&e= )
>>
>> See attached files.
>>
>> My plan was to use this small production run with 0.5 lambda to
>> generate starting structures for other runs and windows, but for most
>> of tried ligands, I am getting them out of pocket in first few
>> picoseconds even at lambda 0.5.
>>
>> I know that both endstates are stable under normal MD for 100 ns
>> without problem. If I perform whole workflow for lambda 0.1 and 0.9 -
>> the production with 0.1 is stable. If I switch ligands and repeat
>> procedure again production with lambda 0.1 is stable, but of course
>> with different ligand as more dominant. I tried to play with number of
>> steps during preparation, different SHAKE settings, using CPU or GPU
>> code and more with no luck.
>>
>> I think I am missing something important :) Any help would be
>> appreciated.
>>
>> Juraj Dobias
>>
>>
>>
>>
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.ambermd.org_mailman_listinfo_amber&d=DwIGaQ&c=ZbgFmJjg4pdtrnL2HUJUDw&r=HwsFjSfOtLupDR-NuCP430rdz1DD2LkJxNM3BsKSjrw&m=ENVyFj6_oD0p_qmj-WymfucA1YgEb6nnnRKPbWF9MfA&s=MouUjibXLfwmg6IZNaxB5YvgoqLdHFSZYKcqRIfZ1wU&e=
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.ambermd.org_mailman_listinfo_amber&d=DwIGaQ&c=ZbgFmJjg4pdtrnL2HUJUDw&r=HwsFjSfOtLupDR-NuCP430rdz1DD2LkJxNM3BsKSjrw&m=ENVyFj6_oD0p_qmj-WymfucA1YgEb6nnnRKPbWF9MfA&s=MouUjibXLfwmg6IZNaxB5YvgoqLdHFSZYKcqRIfZ1wU&e=
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Apr 10 2020 - 05:30:03 PDT
