Gaussian accelerated MD (GaMD)

User discussions
1

GROMACS version:
GROMACS modification:
Hi all,

After a quick search, I see that GaMD is not implemented in GROMACS. Is there any plan to include it in future releases? I found this branch in Github https://github.com/emptyewer/gromacs-gamd (it has 5 years old, though). Does anyone used it and/or recommended it?

Thanks in advance.

Best,
-Yasser

2

Hi Yasser,

In a way it’s already implemented in GROMACS under the name “essential dynamics sampling” - though I have to warn that it is not tested extensively and the code is a bit dated, it might be just what you want! Have a look here

and a corresponding publication here:

Best,
Christian

3

Thanks! I will have a look.

Best.
-Yasser

4

Yes, it is pretty outdated, but I will give it a try. In any case, I think that, if possible, it would be very useful for the community to fully implement an updated version of the method like in NAMD and AMBER (http://miao.compbio.ku.edu/GaMD/).

Best,
-Yasser

5

Hi Yasser,

Rather than implement lots of types of enhanced sampling methods, we are currently focusing on better interaction with packages that provide this functionality, e.g., colvars (https://gitlab.com/gromacs/gromacs/-/issues/3357) and plumed.

The reason behind this is that we would like to make sure that GROMACS is very stable and tested and that becomes harder the more features we incorporate - so we try to split responsibility and provide all hooks and base-layers others can then build upon.

6

Hello Christian, it is worth noting that GaMD is implemented by changing individual potential energy terms, and is not a collective variable-based method. So it would be very difficult to support it in GROMACS only by applying external forces, like Colvars or PLUMED do.

Yasser, why don’t you get in touch with the authors of the GaMD papers and ask them for suggestions?

Giacomo

7

Dear Christian and Giacomo,

Thanks for your replies.

I contacted the author of GaMD, Prof. Yinglong Miao about the method. It is based on a similar mathematical formalism of the bias potential of conformational flooding. However, GaMD smooths the potential energy surface without the requirement of predefined collective variables. Also, it provides rigorous ways to re-weight both thermodynamics and kinetics from the simulations. All the information is on the website in my previous post.

I don’t know about the intricacies of the development process, which is great, btw. But, I do think that the implementation of this method could be useful for the community. Especially to avoid jumping to other software (NAMD, AMBER) and keep the reproducibility. That’s why my question about gromacs-gamd in Github because it looked like an independent development branch.

Best,
-Yasser

8

Dear All,

I was reading through this and can confirm that Yasser is correct, in that GAMD is not a random method but is in fact something that is being more frequently used since 2020.

For example, note this paper where the authors use WT-MetaD coupled to a parallel tempering approach with different values of gamma:
https://pubs.acs.org/doi/10.1021/ct5009087

Note as well the exact same thing here has been done by GAMD and parallel tempering by having consecutive replicas where the potential energy is increasingly elevated:
https://pubs.acs.org/doi/full/10.1021/acs.jctc.4c00501

Similarly, there is a paper here where MetaD methods are used to sample ligand binding:
https://pubs.acs.org/doi/full/10.1021/acs.jcim.6b00772

GAMD being used here to study ligand binding:
https://www.nature.com/articles/s41598-024-58945-4

It is unsurprising that GAMD is now being used to recapitulate a lot of the achievements of MetaD, since restraining potential energy has a similar effect as biasing potential energy as a CV by WT-MetaD until you enter the Well-Tempered ensemble - where the value of gamma affects the fluctuations in potential energy.

I can confirm that GAMD is possible in GROMACS via Plumed but only by using the entire system potential energy as a CV. The idea is to use the system potential energy as the CV for WT-MetaD to generate the well-tempered ensemble for a fixed value of gamma. Both WT-MetaD with system potential energy biased, and GAMD with system potential energy biased, would be more powerful via Plumed if you could select subsets of the system potential energy: e.g. if you could make dihedral angle energies of a protein motif or a ligand the PE term for GAMD. There are then powerful applications for enhancing REST2 by combining it with WT-MetaD or GAMD if you could do this.

See my recommendation here in the Plumed site:
https://groups.google.com/g/plumed-users/c/WLCN5WXkpqg

I agree you will see users switch to AMBER if this is not implemented. I hope I have convinced you that this would be useful to do - you just have to make it such that components or elements of the potential energy function can be selected as an order parameter for biasing, and you would have the implementation fixed for GROMACS. Whether this is done internally or through Plumed is probably something you would all know better.

Kind regards,
Billy