[phenixbb] map deposition strategy after density modification

Wed Sep 20 06:34:03 PDT 2023

Hi Guillaume,

Thanks for the feedback. The previous answers posted here and feedback I've gotten in person and at other online platforms (e.g. Takanori's reply on Twitter https://twitter.com/biochem_fan/status/1599623792040742914) have been very enlightening.

In the relevant case, the density modification did not improve the maps in the regions of interest (even though the best resolved regions looked better in the DM map), so therefore we decided to not use the DM maps and do all the model building, refinement and B-factor assignment against the final, unsharpened map from the 3D reconstruction software.

Regards,
Daniel

From: <phenixbb-bounces at phenix-online.org> on behalf of Guillaume Gaullier <guillaume.gaullier at kemi.uu.se>
Date: Wednesday, 20 September 2023 at 10:59
To: "phenixbb at phenix-online.org" <phenixbb at phenix-online.org>
Subject: Re: [phenixbb] map deposition strategy after density modification

Hi Daniel,

Finally taking time to answer this message (I saved it when it came up; I know we discussed this a few times since then, but I figured I should also post this publicly).

1) I prefer refining against the least "massaged" map, so the one fresh out of the 3D refinement I consider "final". This simplifies many decisions down the line, as you will see in my answers to your other points.
I also use sharpened or otherwise post-processed maps, but only during manual model building/refinement. My favorite tools these days are ISOLDE for manual model building/refinement and deepEMhancer for map post-processing (very easy to use, and almost no parameters to tweak, so less user bias). I use the "final" map to guide MDFF (= pull on the atoms) in ISOLDE, and the deepEMhanced map on top of it, but only used as a visual guide (it doesn’t guide MDFF). Many times, I found that locations with poor or no side-chain density are more easily interpretable in the deepEMhanced map, and that the side-chain conformation suggested by this map ends up nicely stable in the combined pull of the "final" map and the MD force field (without the hint from the deepEMhanced map, it is sometimes a big headache to find such a stable conformation). When I think I can’t improve the model any further in ISOLDE, I use the parameter file it generates for phenix.real_space_refine (this parameter file essentially turns off most things except B-factor refinement) and run a refinement, and the output model is the one I deposit.

2) I always deposit the "final" map (the one fresh out of my final 3D refinement, see 1) as "main map" in the deposition system. I believe this is the most honest thing to do, because it is whatever I saw in this map that convinced me to deposit it and write a paper about it. So, this is the map I want people to see when they simply run "open emdb:xyz" in ChimeraX, even if it’s not the prettiest map and not necessarily the one I use to make figures. I also deposit the deepEMhanced map as "additional map", because I used it to guide model building, and because I most likely used it for the figures (I sometimes use both maps in figures; in any case, I always clearly indicate which map is shown in which figure).

3) I quote the global resolution from the "final" 3D refinement (see 1). It depends on the mask used, so of course I also deposit this mask along with the mandatory half-maps. Unless the particle under study is very rigid and stable, a single global resolution number is pretty much meaningless (in most cases, the location in the reconstruction where the interesting things are has much worse resolution than the global number suggests), so I choose to spend the least amount of time possible coming up with this number. What matters to me is only that people can use the deposited half-maps and mask to recalculate this number, and find something in the same ballpark as I reported.

4) I deposit the FSC curve I obtain from the two half-maps from the "final" 3D refinement (see 1). Same as for the global resolution number: I simply want people to know where this comes from and be able to recalculate it from the deposited data.

5) I deposit the half-maps from the "final" 3D refinement (see 1). I make sure to deposit the mask used in refinement and the mask used to calculate the masked FSC curve (sometimes they differ, for instance cryoSPARC comes up with a tighter mask to give a more optimistic global resolution number because it looks nicer in a title or abstract).

So, you could sum up my approach like this, independently of the programs used: "deposit as main map the least massaged map that convinced me something interesting is here to see; deposit the corresponding half-maps, masks and FSC curve so they agree with this map; deposit as additional map any other map that was helpful along the way; clearly explain in the paper which map was used for what".
I try to go for the simplest possible process from "final" map to deposited files and figures, because I think it is more transparent this way, and I want readers of my papers to see the same evidence that convinced me in the first place to write a paper about it.

I admit this is very opinionated, and mostly originates from my own experience of what was helpful in my hands, but hopefully this is helpful to others as well. :-)
Happy to learn from other opinionated people if their approach differs.

Cheers,

Guillaume

On 2 Dec 2022, at 14:22, Daniel Larsson <daniel.larsson at icm.uu.se<mailto:daniel.larsson at icm.uu.se>> wrote:

Hi Phenix community,

I’ve been using density modification with some success for my data and am considering the most appropriate way to do model building/refinement and then deposition of my results. Here are some questions/observations:

1) Which map should I refine against (most importantly, do the final minimization using phenix.real_space_refine)? The DM map is the most clear one, so I’m favoring to do it against this one (as long as the geometry/validation is also best for this one).

2) Which map should I deposit as my main map? Is it the autosharpened one or the DM one? Since the DM map looks the best, I want the user who only downloads one map to see this one, so again I favoring this map.

3) Which nominal resolution should I quote for my reconstruction? Since the procedure seems to introduce correlation between the two maps (see point 4 below), it does not seem to be possible to obtain a gold-standard FSC value between two truly independent half-maps. As the DM procedure supposedly actually improves the map (by leveraging knowledge about what solvent and macromolecular maps usually looks like and thereby improves the Fourier coefficients), it feels like it would be misrepresenting the data by quoting the FSC_0.143 from the original reconstruction. Should I use the FSC_0.5 estimated value that phenix.resolve_cryo_em outputs?

4) Which FSC curve should I deposit? When I calculate the FSC between the two density modified half-maps, I get a funny (or rather not so funny) looking curve with a bump at around 3Å and it does not go to zero at Nyquist (which does not agree with the map quality). This must again be due to the induced correlation between the maps, presumably during the masking of solvent and macromolecule density. Phenix.resolve_cryo_em calculates a FSC curve to calculate the FSC_0.5 value, which according to the manual page corrects for the correlation induced between the two maps:

"Density modification introduces some correlations between half-maps due to solvent flattening. This can have a small effect on the resolution estimates obtained with half-map FSC. The resolution estimates provided by the program are corrected for this effect."

But is this curve good enough? I think OneDeposit wants to have the gold-standard FSC curve, right?

5) Which (nowadays mandatory) half-maps should one deposit? Here I would say that the original half-maps before DM makes more sense, since these are less processed. But if one specify these as the half-maps, there will be inconsistencies during validation, since the DM map will look quite different from the average between the two original half-maps. Also the automatic FSC curve will be very different from the deposited one.

So in conclusion, despite what I say above, the most conservative thing would be to deposit the conventionally sharpened map as my main map and use the original half-maps for calculating the FSC-curve and use the 0.143 threshold from this curve to estimate the resolution. The DM map can be used during rebuilding/refinement, but I think final minimization should be against the main map during deposition (with restraints if the geometry gets worse). The DM map can be deposited as an auxiliary map.

Do the community and the developers agree?

Regards,
Daniel

När du har kontakt med oss på Uppsala universitet med e-post så innebär det att vi behandlar dina personuppgifter. För att läsa mer om hur vi gör det kan du läsa här: http://www.uu.se/om-uu/dataskydd-personuppgifter/

E-mailing Uppsala University means that we will process your personal data. For more information on how this is performed, please read here: http://www.uu.se/en/about-uu/data-protection-policy

_______________________________________________
phenixbb mailing list
phenixbb at phenix-online.org<mailto:phenixbb at phenix-online.org>
http://phenix-online.org/mailman/listinfo/phenixbb
Unsubscribe: phenixbb-leave at phenix-online.org

VARNING: Klicka inte på länkar och öppna inte bilagor om du inte känner igen avsändaren och vet att innehållet är säkert.
CAUTION: Do not click on links or open attachments unless you recognise the sender and know the content is safe.
_______________________________________________ phenixbb mailing list phenixbb at phenix-online.org http://phenix-online.org/mailman/listinfo/phenixbb Unsubscribe: phenixbb-leave at phenix-online.org
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20230920/5f2495ed/attachment.htm>