protein_specific_ffield

[karolk]

Hi,

I would like to design my own protein-specific ffield. What would be the best strategy here? How I should start? Any references will be appreciated.

I found that you can build custom functions in bio3d, but I am not sure whether this is what I want.

Any comments will be greatly appreciated thanks!

[Lars Skjærven]

the bio3d NMA functions are designed to easily define custom "force fields". see some examples in bio3d/R/load.enmff.R. you need some clear ideas on WHY and HOW for how to proceed with this. You can start with Hinsen et al 2000, or check out the series of publications on the REACH force field (Moritsugu and Smith), as well as the Dehouck and Mikhailov 2013.

[karolk]

Hi,

Actually, I have already read those.

The reason why I want to define my ffield is that the results of my NMA analysis are very ffield dependent, which (I think) should not be like that. There are certain cases (ffields), where results are completely wrong and do not look as they should at all = NMA fails. The most reliable results yield these ffields, which should be the least-protein(shape) dependent, that is, pfNMA, REACH and HCA by Hinsen. Thats way I have been thinking that it would be wise to check, whether I can get results, which would be more reliable or comparable to the ones, which were obtained with the aforementioned ffields.

And what is really unexpected is the fact that All-atom NMA with -rtb on, yielded worst results.

thanks

[Xinqiu Yao]

When you say "worst" or "wrong", what values are you talking about, fluctuation, covariance, or something else? And how did you benchmark the results?

[karolk]

By worst I mean:. with the exception of a few small regions, the protein is completely static. No fluctuations at all. I can reference this to experimental data, so, I am pretty sure that it is wrong. Additionally, I compare RTB to results obtained with various ENM force-fields. The ENM is in much better agreement with this what I see in experiment. Still, based on this what I read, RTB should not fail so badly here. Quite opposite, it should be one of the best ones. I tried nmer 1 and 6 with and without reduced option.

[Lars Skjærven]

Note that you should not over interpret the amplitudes from NMA. When you say that your protein is "completely static", you probably mean that the fluctuation amplitudes are low? To evaluate the performance of NMA you should look at directionalities and not amplitudes. see e.g. the rmsip and bhattacharyya functions (http://thegrantlab.org/bio3d/tutorials/ensemble-nma-part-1), to compare e.g. with PCA or MD.

[karolk]

Yes. I know. By static I did not mean that amplitudes are low or lower than...I meant, with the exception of a few small regions, protein does not move at all, and this is not normal.

[Xinqiu Yao]

Maybe it is too much for you, but could you perform a NMA with e.g. Amber force field and see what it looks like? This should be the most accurate model and if it is similar to aanma, it might be a intrinsic property of the structure. Note that I am not saying the protein actually doesn't move because you indeed saw fluctuations from experiment. It means NMA (whether Amber or aanma based) might not capture those fluctuations because of the rugged energy landscape.

Another thing you can try is testing it on an alternative structure (if available) or one after energy minimization. That might change the topology and so results.

The problem is your system is large and so difficulty for us to test and reproduce your problem. But let me know what you think about these suggestions and see what I can do to help.

[karolk]

Hey,

Thank you for your input. Very helpful.

I have been also thinking about doing emin prior to NMA. I assume that you minimize until you reach plateau? The energy does not change anymore. Right?

I did tests on alternative structures as well and they all show the same, at least using Hinsen Calpha and REACH ffields. I did not try aa-nma with/without RTB. I will try it out.

What do you mean by "could you perform a NMA with e.g. Amber force field and see what it looks like?" You mean: Calpha ffield by Hinsen? I did it and all structures show what is expected, but the all-atom representation contain some problematic regions, so I will first try to minimize.

I will also include ligands. Their presence might make a difference here.

thanks

[Xinqiu Yao]

By "Amber force field", I mean to perform NMA on the potential energy defined by Amber force field (such ff14sb+gb/sa). Either Hinsen's C-alpha force field or the one from aanma() is a kind of "coarse-grained" model called elastic network model, which neglects detail about atomic interactions such as electrostatics and Van der Waals. You can find descriptions about NMA using Amber force field from the Amber manual http://ambermd.org/doc12/Amber17.pdf, or our recent paper http://pubs.acs.org/doi/10.1021/acs.jpcb.6b01991.

During Amber-based NMA, you will get an energy-minimized structure. You can then use that structure to test ENM-based NMA. It is similar to regular EM for MD but without water (instead we used GB/SA) and stricter about convergence (See above references for more detail).

Let me know if you have additional problems.