search

leelasd / ligpargen

A repository for tutorials and FAQ's about LigParGen
http://traken.chem.yale.edu/ligpargen/
21 stars 17 forks source link

Parameters for highly charged molecules #10

Open matteoaldeghi opened 6 years ago

matteoaldeghi commented 6 years ago

I wonder whether it is possible to allow molecules with net charge larger than +/-2 to be parameterised by the LigParGen Server?

leelasd commented 6 years ago

@matteoaldeghi Problem with AM1 semi-empirical model and CM1A charge model is that they not parameterized for charges other than +2,-2,+1,-2,0. So, I suspect how well they will perform for highly charged systems.

matteoaldeghi commented 6 years ago

@leelasd I see the problem - allowing only certain charges, however, really limits the usability of the force field in particular when it comes to simulating things like cofactors (e.g. ATP, GTP) that are quite common. Maybe an option could be to allow charges above 2 but at the same time give a big warning about the fact the charge model was not parameterised for such systems?

I also noticed that you now provide a way to install and run ligpargen locally via command line, which is very useful. Is the charge limitation still present also when using the program via command line? My practical issue is that I am testing/comparing a few force fields for free energy calculations using Gromacs, and I would like to include OPLS too, but many of the molecules I am looking at are highly charged...

leelasd commented 6 years ago

@matteoaldeghi I agree with you, It would be useful to allow higher charge values but it's important to realize that molecules like ATP, GTP and other nucleotides need to be parameterized explicitly. In fact, one of my coworkers published papers on nucleotide parameters. We don't really want people to use charges that are not accurate from the web server, that makes the force field look bad. I can enable the option of higher charges for local version LigParGen (assuming they are aware of the limitations)

Regarding Local version of LigParGen, can you email Prof. Jorgensen and cc me, requesting BOSS software. Then I can send you everything you need for setting up a local version of LigParGen. Also, It would be an unfair comparison if you are comparing FF for these highly charged systems, most of them may not have been included in parameterization set 😃 My guess is most existing FF parameters are poor for these systems.

matteoaldeghi commented 6 years ago

@leelasd firstly, thanks once again for your help/suggestions!

It would be useful to allow higher charge values but it's important to realize that molecules like ATP, GTP and other nucleotides need to be parameterized explicitly. In fact, one of my coworkers published papers on nucleotide parameters.

I found the paper by Michael you mentioned. I imagine the force field files are not yet available for Gromacs? Another option would be to introduce those updated torsions for phosphates etc. in ligpargen directly? I believe the cgenff program does something similar, if a nucleotide is provided, the best available Charmm parameters are assigned. Btw, it is just a suggestion off the top of my head - I realise I keep suggesting you guys more work to do while you probably have better things to take care of!

I can enable the option of higher charges for local version LigParGen (assuming they are aware of the limitations) Regarding Local version of LigParGen, can you email Prof. Jorgensen and cc me, requesting BOSS software. Then I can send you everything you need for setting up a local version of LigParGen.

This would be great! I will email Prof. Jorgensen soon.

Also, It would be an unfair comparison if you are comparing FF for these highly charged systems, most of them may not have been included in parameterization set 😃 My guess is most existing FF parameters are poor for these systems.

I totally agree that the dataset I am using is quite a challenging one when it comes to force fields; the test set I am using is really driven by what is available experimentally, and just so happened that some of the cases I am looking at are quite challenging (and as you say this will be the case for most existing force fields). That said, only a bit less than 1/3 of the ~25 ligands I am looking at are highly charged, the others are more standard synthetic compounds. Just to clarify a bit more what I am doing: I am benchmarking our free energy protocol for protein, rather than ligand, mutations (i.e. see how the ligand affinity changes upon protein mutation). The ligand will still have an effect, but maybe not as large as when doing ligand free energy calculations. Also, in many cases the highly charged molecule (like ATP) is just the cofactor that needs to be present but it is not really directly involved in the contact with the mutated residue. My interest in the comparison is to see how different force fields do for this particular problem when using currently available ff parameters (which might indeed not be particularly good for some of the more tough systems) so to have a general idea about which ff to move forward with.

Btw I imagine indeed that for instance the AM1-BCC charges in GAFF would have similar issues. However, in Amber there is also the option to move to RESP. Is there a similar option for OPLS, with somewhat automated tools? If this is a more appropriate way to deal with large ligand charges, I would be happy to try go down this road.