Re: [AMBER] tleap/xleap performance enhancement

From: Matias Machado <>
Date: Mon, 10 Feb 2020 12:35:05 -0300 (UYT)

Dear Himanshu,

Dealing with huge systems is a challenging issue I would like to address as well...

These are some tips that may help (or may not)...

1) "The most time consuming process is when it prints "Starting new chain with *segname*". Will it help if we somehow disable the print commands."

+++ To avoid this behaviour remove SEGID field from your input PDB. I'm not a developer, but leap seems to process that field for some reason, however the generated topology is identical with or without SEGID.

+++ In addition, you can also save "some" I/O processing time by redirecting the standard output to a file or /dev/null, e.g.:

tleap -f input &> /dev/null ;# avoid using xleap (GUI), rendering to x11 may be time consuming.

+++ You can avoid creating a log file by commenting/removing the following line from cmd files "leaprc.protein.ff14SB" and "leaprc.DNA.bsc1"

logFile leap.log

2) "the tleap proceeding becomes even slower as the time proceeds"

+++ This smells to me a memory (RAM) issue or a nested loop issue...

3) "I stripped hydrogen and kept only oxygen atoms in the pdb file before loading"

+++ I see no gain in removing hydrogen atoms from water molecules, in doing so you are just vanishing the hydrogen-bond network, hence, which is the advantage of using a pre-equilibrated solvent?

4) "leap takes ~1 hour to read the pdb and create the prmtop and inpcrd"

+++ I strongly recommend you to use NetCDF format for coordinates, which is unrestricted to system size, lighter, more efficient and accurate for computing than inpcrd (ASCII), in particular for such huge system you are building...



Researcher at Biomolecular Simulations Lab.
Institut Pasteur de Montevideo | Uruguay

----- Mensaje original -----
De: "Himanshu Joshi" <>
Para: "david case" <>, "AMBER Mailing List" <>
Enviados: Viernes, 7 de Febrero 2020 17:12:44
Asunto: Re: [AMBER] tleap/xleap performance enhancement

What I am most intrigued here is, I have ~4 million atoms of solute and
leap takes ~1 hour to read the pdb and create the prmtop and inpcrd files
Whereas for the similar number of water molecules, leap is ~50 times slower
in creating the input files.

Note: I stripped hydrogen and kept only oxygen atoms in the pdb file before

ATOM ***** O WAT 7160745 -179.864 8.554 -13.358 1.00 0.00 A
ATOM ***** O WAT 7160746 37.736-113.528-227.411 1.00 0.00 A

Am I missing something basic here.


On Fri, Feb 7, 2020 at 11:08 AM Himanshu Joshi <>

> Dear Prof Case,
> Thanks for your response, I will appreciate if you can suggest any
> possible get around.
> There are solutes like DNA, protein and counterions in the system, I am
> using parmbsc1 and ff14 for the DNA and protein, tip3p for water model.
> However, the time consuming part is the water in the system,
> One update, the tleap proceeding becomes even slower as the time
> proceeds, contrary to earlier estimate, after 12 hours it has created only
> 500,000 water
> molecules. And now I am concerned that it might be slower towards the
> end.
> Thank you.
> Sincerely
> Himanshu
> On Fri, Feb 7, 2020 at 7:00 AM David A Case <>
> wrote:
>> On Thu, Feb 06, 2020, Himanshu Joshi wrote:
>> >
>> >Although, I could manage to tweak the pdb file format to get recognized
>> by
>> >tleap but the process of generating the topology and coordinate file is
>> >slower, (approximately 12 hours for 1 million atoms).
>> What force fields are you using? Is this just pure water, or are there
>> solutes?
>> I've seen behavior that sounds similar, but I'm trying to narrow down
>> when it happens. No promises, however, that there will be any easy fix
>> or workaround.
>> ...dac
>> _______________________________________________
>> AMBER mailing list
> --
> *With Regards,HIMANSHU JOSHI *

*With Regards,HIMANSHU JOSHI Graduate Scholar, Center for Condense Matter
TheoryDepartment of Physics IISc.,Bangalore India 560012*
AMBER mailing list
AMBER mailing list
Received on Mon Feb 10 2020 - 08:00:03 PST
Custom Search