Best practices for MMPBSA computation from MD

[Vincent Binette]

Dear all,

I am currently computing the binding affinity ligand/protein from a single trajectory MD simulation. I compared the results computed from autodock VINA, your code (gmx_mmpbsa) and g_mmpbsa (see attachment). Those results are computed on the interval of convergence. 

With gmx_mmpbsa, the values are high with very large fluctuations compared to, let's say, g_mmpbsa (mostly coming from EEL and EPB).

I was wondering if you had any recommendation for MMPBSA computation. Is it better to minimized all structures before computation? Should I consider the hydrogen atoms? Do you have any thoughts on why the results are so vastly different?

 Best regards,

Vincent Binette

[Mario Ernesto]

Hi Vincent!

Thanks for your interest in our tool. The results you described are totally expected since the three methods you are using are different. Vine uses and scoring function, g_mmpbsa uses APBS while gmx_MMPBSA (based on AMBER's MMPBSA.py) is using the built-in internal PBSA solver in AMBER. It is possible to get similar energies between g_mmbsa and gmx_MMPBSA if you use APBS in sander. The thing is you need to install APBS for amber (https://mccammon.ucsd.edu/iapbs/doc/amber.html#_building_of_amber_apbs_module). That way you can set the same parameters for both programs (g_mmpbsa and gmx_MMPBSA) when running a PB calculation. Make sure to set the variable `sander_apbs`=1 once you install sander APBS. We should go over this process in the near future and make the installation process easier for the user.

Let us know if you need any further information...

Best!

 

--
You received this message because you are subscribed to the Google Groups "gmx_MMPBSA" group.
To unsubscribe from this group and stop receiving emails from it, send an email to gmx_mmpbsa+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/gmx_mmpbsa/f05364a1-70ba-4d55-b46d-8827743eb7f9n%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Vincent Binette]

Dear sir,

Thank you for the quick answer.

If I understand correctly, the results between gmx_mmpbsa and g_mmpbsa being highly different (with a low correlation) is expected based on the different methodology used. I am still unsure what method may be better suited for my system. From what I have seen in the litterature, the justification behind the choice of MMPBSA computation method is not something being really discussed in articles.

In your experience, on what criteria do people choose their MMPBSA methodology in the litterature? Would you have an idea of what kind of systems might be better suited for g_mmpbsa or gmx_mmpbsa? Or maybe the choice is necessitated in order to compare with data in the litterature (my guess is that using the same computation method is best for comparison)?

Best regards,

Vincent Binette

[Mario Ernesto]

>If I understand correctly, the results between gmx_mmpbsa and g_mmpbsa being highly different (with a low correlation) is expected based on the different methodology used.

Yes, this is correct

>I am still unsure what method may be better suited for my system. From what I have seen in the litterature, the justification behind the choice of MMPBSA computation method is not something being really discussed in articles.

End point free energy calculation methods (like MMPBGBSA) are suitable for calculating the relative binding free energy differences with a low computational cost. If I understand correctly, you are working with a single system, that way MMPBGBSA perhaps is not the way to go since it's not going to give you meaningful results when calculating absolute energies... I'll suggest LIE (Linear interaction energy) method (this is also an end point free energy calculation method) or something more fancy like TI, FEP or umbrella sampling if you have enough computational power to do so.

>Would you have an idea of what kind of systems might be better suited for g_mmpbsa or gmx_mmpbsa?

In principle, you can handle a number of systems with gmx_mmpbsa (protein-protein, protein-DNA, etc). Usually, large systems that are difficult to handle with more accurate methods like TI or FEP

>Or maybe the choice is necessitated in order to compare with data in the litterature (my guess is that using the same computation method is best for comparison)?

This is the case sometimes... others, as I said, it's more related to the system itself... large systems are difficult to handle with more accurate methods and the results are not that good sometimes...

hope this helps!

best!

To view this discussion on the web visit https://groups.google.com/d/msgid/gmx_mmpbsa/d1efe3b0-cac5-45c1-b8da-d13bb978d05dn%40googlegroups.com.