ChimeraX

[TBGAnsell]

Do you have plans to implement usage in ChimeraX?

[gregdp]

Thanks for the question and your interest. Yes I am moving it to ChimeraX, hope to have something for it soon. Could always use some help in that regard if anyone has experience with python and ChimeraX ;)

[ctueting]

Any news regarding moving this to ChimeraX?

[kvr2007]

A plugin is now available for ChimeraX:

https://github.com/tristanic/chimerax-qscore

You can also get it from the toolshed:

https://cxtoolshed.rbvi.ucsf.edu/apps/chimeraxqscore

So I guess the issue can be closed :)

[kuixu]

I tested PDB-6cvm and emd-7770 both in Chimera and ChimeraX. But the Q-score is different.

In Chimera, the calculated value is 0.70 (the same as the validation report on RCSB PDB), but in ChimeraX, the calculated value is 0.73 (sigma=0.4).
The sigma setting is 0.4 in both ChimeraX and Chimera.

So, why are the Q-scores calculated in ChimeraX different from those in Chimera?

[gregdp]

The ChimeraX implementation is indeed different, particularly in how points are distributed around each atom.

• The Chimera implementation aims to make them evenly spaced, while maintaining the constraint that they are closer to the atom of interest rather than any other atom. It is a slower computation, due to this 'nearest point' search being performed more times. I think with the points more evenly distributed, it should be a slightly more reliable Q-score.
• In ChimeraX a more approximate but faster k-means approach is used - starting with more points, and reducing them by clustering, done in C++ which is also faster than Python. With this approach, the points should end up being evenly distributed in general, but there may be cases where they are more clumped together perhaps, not capturing the density profile as accurately.

I have noticed results can be different, maybe this has to be stated a bit more clearly somewhere. In some of my tests I saw they can vary on average by ~1%. In this case it seems a bit more drastic, being around 5% difference. I think in general they are still quite close and indicative of resolvability. I hope there aren't cases where you would get something like .7 in Chimera and .3 in ChimeraX, for example. So I don't think anyone should hold back from using ChimeraX version, again due to the difference not being that large, it is still very useful in my opinion for catching problems in the model and map. If a larger difference is found, I would be very interested to know.

As a side note, the EMDB is calculating Q-scores with the Chimera plugin, so if you are comparing with the EMDB values, you may also notice this difference.