Measure fit of residue in map

[tic20@…]

Begin forwarded message:

From: Tristan Croll
Subject: Re: ISOLDE and ChimeraX plan for the future
Date: May 4, 2020 at 3:24:17 PM PDT
To: Tom Goddard

Hi Tom,

...

I can also think of a few map-related tools that would be useful. One that would be *really* helpful is a local fit-to-density measure, to flag residues that aren't currently well fitted to the map. Various approaches to that, of course - one of which would be to rejuvenate/reincarnate the old TEMPy plugin. I had a chat to Maya about this last year - she's all for it, but the person who wrote the original has long since moved on, and she doesn't currently have the resources to pick it back up.

...

[Tom Goddard]

The hard part of this is how to make it usable. There may be 25% of residues in poorly resolved cryoEM density that would all be flagged as poor fits. I guess the dream is some tool that says this residue fit looks much worse than expected for this local density resolution and signal to noise.

[Tristan Croll]

This is indeed a somewhat difficult problem - which the official wwPDB 
cryo-EM validation report currently suffers badly from. It just takes 
the author-recommended contour level, then reports on the number of 
atoms outside that contour (and number of residues with >50% of atoms 
outside). It's a terrible metric that (amongst other things) causes 
problems with reviewers: for a typical cryo-EM map with some 
well-resolved regions and some poor, picking a contour that's right for 
the good regions leaves a lot of the poorly-resolved residues outside, 
and a naive reviewer will then decide this means the model must be crap.

I understand this is a fairly active area of research. I saw a talk 
online from last week's CCP-EM Spring Symposium from Ardan Patwardhan 
(https://www.ebi.ac.uk/about/people/ardan-patwardhan) describing what he 
called a "STRUDEL score" - if I understood correctly, checking the 
correlation of real-world examples of residue density at similar 
resolution to the density around each model in the actual residue, which 
made the sequence error in the SARS-CoV2 RNA polymerase stick out like a 
sore thumb. He said in his talk that there's a ChimeraX plugin for this 
ready to go - have you heard anything about it?

More generally, a good approach would probably be to weight any such 
scoring system against a local-resolution estimation (another thing that 
would be nice to have, come to think of it - will raise another ticket).

On 2020-05-05 00:57, ChimeraX wrote:

[goddard@…]

It does seem like a local resolution calculation would help in deciding if a residue fit is poor.  Ticket #3144 is about adding local resolution calculation.