Re: ParseFEP error with FEP in segments at different lambda schedule

From: Francesco Pietra (chiendarret_at_gmail.com)
Date: Sat Aug 04 2018 - 08:21:10 CDT

To those who have followed this thread, if any:

the Sterr can be calculated from the Stdev and the sample size, directly
from the online calculator
https://mathcracker.com/standard-error-calculator.php#results , whereby,
from the Stdev that I reported, 91.0, and the sample size 6204, the Sterr
is 1.15

I submit that the Sterr is more representative that the Stdev in this
context.

francesco pietra

On Wed, Aug 1, 2018 at 4:10 PM Francesco Pietra <chiendarret_at_gmail.com>
wrote:

> Standard deviation, as provided by xmgrace when loading the .agr file
> created with NAMD plot.
>
> Could xmgrace, or other software, be turned to calculate the standard
> error? I am dealing with very long trajectories, so that hand calculation
> is hardly feasible. On the other hand, in comparison of different poses for
> the same ligand and receptor, MM-GBSA could be of some utility. Otherwise,
> could you provide a link to how implementing entropy with MM-GBSA?
>
> I turned to MM-GBSA while attempting to improve on absolute FEP. Nearly
> perfect frwd+back "probability" superimposition proved not enough: I
> obtained large scatter of /\A for the 0.8-1.0 sector (the one on which I am
> now focusing the attention) on change of flags that should be irrelevant if
> quasi converge had been reached. I am now waiting results for FEP at the
> limit of what can be done in a 24hr run for 0.8-1.0 on my cluster (number
> of windows/number of FEP steps, ts=2fs, while I used before ts=1.0 fs). For
> a system of biological interest I am faced by such a high wall. In other
> quarters, absolute FEP is carried out on GPUs (none of them on my cluster),
> so that the "wall" should be much lower than with CPUs. But, was the
> CPU/GPU comparison well tested?
>
> Thanks for advice
> francesco
>
> On Wed, Aug 1, 2018 at 1:53 PM Brian Radak <brian.radak_at_gmail.com> wrote:
>
>> Sorry, yes you are correct, that should be "sum of the square of the
>> errors".
>>
>> Are you using the standard deviation or the standard error? The latter is
>> more difficult to calculate but almost certainly smaller (unless your
>> effective sample size is abysmal). Nonetheless, I believe this has always
>> been considered one of the drawbacks to MM/GBSA-type methods. That and the
>> accurate inclusion (or lack thereof) of entropic effects.
>>
>> On Wed, Aug 1, 2018 at 6:40 AM, Francesco Pietra <chiendarret_at_gmail.com>
>> wrote:
>>
>>> "the error is square additive, however, so you would take the square
>>> root of the sum of the errors when re-combining"
>>>
>>> May I ask whether the above "sum of the errors" is a shortage for the
>>> "sum of the square of the errors" ? If so, translating to a different
>>> albeit related case, let me show some data (kcal/mol) from MM-GBSA rather
>>> than FEP for the same system, so that /\A below should be understood as
>>> /\POTENTIAL from NAMD plot/xmgrace
>>>
>>> complex: -6335.08 with Stdev 64.24
>>>
>>> receptor: -6358.87 with Stdev 63.93
>>>
>>> ligand: 48.33 with Stdev 6.2
>>>
>>>
>>> /\Abind = complex - receptor - ligand = -24.5 with Stdev 91
>>>
>>> if the propagation of errors is treated here correctly, the whole is
>>> quite disappointing.
>>>
>>> Thanks a lot for checking
>>>
>>> francesco
>>>
>>>
>>>
>>>
>>> On Tue, Apr 10, 2018 at 2:56 PM Brian Radak <brian.radak_at_gmail.com>
>>> wrote:
>>>
>>>> The simple solution to this is to break the simulations into two groups
>>>> where the spacings are identical. So long as you get the directions correct
>>>> the net free energy is just the sum of the two. Most people probably know
>>>> this, but just to be complete, the error is square additive, however, so
>>>> you would take the square root of the sum of the errors when re-combining.
>>>>
>>>> HTH,
>>>> BKR
>>>>
>>>> On Sun, Apr 8, 2018 at 9:58 AM, Chris Chipot <chipot_at_ks.uiuc.edu>
>>>> wrote:
>>>>
>>>>> Francesco,
>>>>>
>>>>> ParseFEP assumes constant ∆λ throughout the free-energy calculations.
>>>>>
>>>>> Cheers,
>>>>>
>>>>> Chris Chipot
>>>>>
>>>>>
>>>>>
>>>>> On 4/8/18 3:46 PM, Francesco Pietra wrote:
>>>>>
>>>>> Hello:
>>>>> I carried out FEP for Unbound ligand in water along two segments, as
>>>>> follows:
>>>>>
>>>>> frwd-01
>>>>> 0.00 0.20 0.05 $numSteps
>>>>>
>>>>> frwd-02
>>>>> 0.20 1.00 0.10 $numSteps
>>>>>
>>>>> back-01
>>>>> 1.00 0.80 -0.05 $numSteps
>>>>>
>>>>> back-02
>>>>> 0.80 0.00 -0.10 $numSteps
>>>>>
>>>>> concatenating the results as follows:
>>>>> frwd-01.fepout > frwd.fepout
>>>>> frwd-02.fepout >> frwd.fepout
>>>>>
>>>>> back-01.fepout > back.fepout
>>>>> back-02.fepout >> back.fepout
>>>>>
>>>>> ParseFEP with SOS estimator gave error:
>>>>>
>>>>>> domain error: argument not in valid range
>>>>>> domain error: argument not in valid range
>>>>>> while executing
>>>>>> "expr {$mean/$nfepdata}"
>>>>>> (procedure "::ParseFEP::analysis_normal_result" line 5)
>>>>>> invoked from within
>>>>>> "::ParseFEP::analysis_normal_result $window $fororback $nfepdata
>>>>>> $fepdata"
>>>>>> (procedure "::ParseFEP::normal_parse_log" line 39)
>>>>>> invoked from within
>>>>>> "::ParseFEP::normal_parse_log $::ParseFEP::fepbofile backward "
>>>>>> (procedure "namdparse" line 36)
>>>>>> invoked from within
>>>>>> "namdparse"
>>>>>> (in namespace inscope "::ParseFEP" script line 16)
>>>>>> invoked from within
>>>>>> "::namespace inscope ::ParseFEP {
>>>>>>
>>>>>> if { [string length $::ParseFEP::fepofile] < 1 && [string
>>>>>> length $::ParseFEP::fepdwfile] < 1} {
>>>>>> ..."
>>>>>> invoked from within
>>>>>> ".parseFEP.runbutton invoke"
>>>>>> ("uplevel" body line 1)
>>>>>> invoked from within
>>>>>> "uplevel #0 [list $w invoke]"
>>>>>> (procedure "tk::ButtonUp" line 22)
>>>>>> invoked from within
>>>>>> "tk::ButtonUp .parseFEP.runbutton"
>>>>>> (command bound to event)
>>>>>>
>>>>>
>>>>> ParseFEP with BAR gave no error, but only partial output and no
>>>>> graphics.
>>>>>
>>>>> If I understand, the lambda schedule that I choose is incorrect for
>>>>> the plugin.
>>>>>
>>>>> Previously I carried out successfully the same FEP along four
>>>>> sectors, with same /\lambda throughout. Now, in view of heavy FEP with the
>>>>> complex ligand-protein, I wanted to use tight /\lambda for the steep
>>>>> regions only, apparently with a wrong protocol.
>>>>>
>>>>> Thanks for advice
>>>>>
>>>>> francesco pietra
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> --
>>>>>
>>>>> Chris Chipot
>>>>> CNRS research director, University of Lorraine
>>>>> Adjunct professor of physics, University of Illinois, Urbana-Champaign
>>>>>
>>>>> _______________________________________________________________________
>>>>>
>>>>> Chris Chipot, Ph.D.
>>>>> Theoretical and Computational Biophysics Group
>>>>> 3165 Beckman Institute for Advanced Science and Technology
>>>>> University of Illinois at Urbana-Champaign
>>>>> 405 North Mathews
>>>>> Urbana, Illinois 61801 Phone: (217) 300-0380
>>>>>
>>>>> E-mail: chipot_at_ks.uiuc.edu
>>>>> Christophe.Chipot_at_Univ-Lorraine.fr
>>>>> Web: http://www.ks.uiuc.edu/~chipot
>>>>> http://www.lia-uiuc.cnrs.fr
>>>>>
>>>>> It is hard for France to maintain a thirty-five-hour workweek when
>>>>> China and India have invented a thirty-five-hour workday.
>>>>> Thomas L. Friedman
>>>>> _______________________________________________________________________
>>>>>
>>>>>
>>>>
>>

This archive was generated by hypermail 2.1.6 : Tue Dec 31 2019 - 23:20:09 CST