[phenixbb] rotamer outlier and clash socre

Tristan Croll tic20 at cam.ac.uk
Sun Feb 14 07:17:59 PST 2021

Dear Pan,

I apologise if the following comes across as overly harsh, but this is something that needs to be said (to many in structural biology, not just you):

If you skip the manual component of model building and rely on automated building/refinement tools combined with summary validation statistics to do the job for you, it's a near certainty that the model you deposit will end up with preventable errors. The only real questions are: (a) how many, and (b) how serious. While hopefully we will eventually get to the point where automated building and refinement find the right answer almost all the time, and validation near-infallibly tells us which places are correct and don't need to be manually checked... we're not there yet. Right now, automated rebuilding tends to be too "good" at finding conformations that aren't Ramachandran and rotamer outliers and don't clash (much), but are still wrong. The wwPDB is full of such depositions... here's an example from literally the first (very recent) structure I looked at today. I won't name it (that's not my point), but it's a 3.4 angstrom cryo-EM map with zero Ramachandran and rotamer outliers, and a clashscore of 6. While overall it seems quite good, the pictured site is modelled as in bad_trans_pro.jpg - while there are no Rama/rota outliers, that's only because the geometry's been force-fit into a chemically-infeasible conformation with two amides pointing at each other (which would undoubtedly be flagged as a major clash in MolProbity). Remodelling to a cis proline gives a much more sensible solution, and as a bonus fits the density rather better (both in the backbone and the sidechain of the preceding Phe).

The point of my mini-rant: your role in the process shouldn't be to simply press buttons to start each successive automated task. If you want a good model, you need to get in there and diagnose, on a local level, why these outliers and clashes are appearing. It's quite likely that most of them will be fairly  innocuous and easy to address, but unless you check for yourself you can't rule out the potential for more serious problems.

In terms of tools for this sort of interactive inspection and rebuilding, have you tried ISOLDE (https://isolde.cimr.cam.ac.uk)? This sort of moderate-to-low resolution regime is exactly what I designed it to handle.

Best regards,

From: phenixbb-bounces at phenix-online.org <phenixbb-bounces at phenix-online.org> on behalf of xiaowp82 at hotmail.com <xiaowp82 at hotmail.com>
Sent: 14 February 2021 06:23
To: phenixbb <phenixbb at phenix-online.org>
Subject: [phenixbb] rotamer outlier and clash socre

Hello everyone

I refined a cryo-EM data with 3.2 angstrom, the model fit well with the map, but the rotamer outlier is too high, about 8%.  How can I decrease the value of rotamer outlier by realspace.refinement in Phenix GUI, and no need to adjust the residues one by one in Coot?

Another question is the value of clash score are also too high. When I set the nonbonded_weight to 10000, the clash score get better, but the value of rotamer outlier get worse. The high clash score maybe due to including hydrogens for refinement? Can I exclude hydrogen for refinement and how to do?

Thanks !


xiaowp82 at hotmail.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20210214/3be39c8c/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: bad_trans_pro.jpg
Type: image/jpeg
Size: 31286 bytes
Desc: bad_trans_pro.jpg
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20210214/3be39c8c/attachment-0002.jpg>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: corrected_cis_pro.jpg
Type: image/jpeg
Size: 32224 bytes
Desc: corrected_cis_pro.jpg
URL: <http://phenix-online.org/pipermail/phenixbb/attachments/20210214/3be39c8c/attachment-0003.jpg>

More information about the phenixbb mailing list