[phenixbb] questions/problems with tls_as_xyz routines

Pavel Afonine pafonine at lbl.gov
Mon Apr 18 17:02:05 PDT 2016

Hi David,

> 1. In the paper describing this, Urzhumtsev et al, Acta Cryst D71:1668, 2015,
> there seems to be something wrong with Section 6.2, which describes results
> on PDB entry "1dqv", syanptotagmin III.  The description says that there are
> two copies of the molecule in the asymmetric unit, and the resolution is 2.5
> Ang.
> However, the file with this code in the RCSB (which is indeed syanptotagmin
> III) has one molecule per asymmetric unit, a single chain of 296 amino acids,
> rather than two chains of 97 amino acids, as in Tables 2 and 3 of the paper
> cited above; the resolution is 3.2 Ang.  And it doesn't look anything like
> the picture of "1dqv" in Fig. 3a.
> So something looks wrong; does anyone know what system Section 6.2 is using?

thanks for pointing this out! In fact this is the model of rnase-s (1RGE 
can be used as a reference) and the data we used are one of the sets 
used to solve the structure. I can send you files off-list. This is an 
unfortunate error and we are planning to submit a corrigendum.
1DQV was in a short list of examples we planned to use but in the end 
did not include it.

> 2. I'm asking the above because I'm having trouble with my own applications,
> and was hoping to check that I could reproduce the results in the paper.
> Apropos of that, is there a test case or example for tls_as_xyz in the phenix
> distribution?  [I'm not all that familiar with phenix, so apologies if I've
> missed the easy way to find this.]

We have plenty of unit tests that exercise the math described in the 

Plus of course we have data and model files and calculation results for 
the examples shown in the paper (these are not shipped with Phenix but 
we can provide on request).

> 3. Here is a summary of what is not working for me.  (I don't expect anyone to
> debug this on such little information, but thought that someone on the list
> might have done some similar calculation.)
>      a. I took a small domain (residues 6-61 from GB3, PDB code 2IGD), and
>         used Ethan Merrit's lsq_tls_fit to obtain a set of TLS tensors that
>         best fit the deposited ADP's.  The statisitcs are pretty good,
>         indicating (not surprisingly) that motions in this domain are well
>         characterized as a rigid body--which is why I chose this system.
>         I just used the CA atoms for this initial study.
>      b. I used phenix.tls_as_xyz to compute an ensemble from the TLS matrices
>         obtained in step a.
>      c. I then computed ADP's from the ensemble, using tools from cpptraj
>         in Amber, (or our own codes, which give the same results).  But
>         the extracted isotropic B factors (and the anisotropic
>         components) are much larger than the input ones.  So, the ensemble is
>         giving me what I expected: a sampling of structures whose fluctuations
>         arise from the input TLS matrices.
> I've done a fair amount of due diligence in trying to track down errors on my
> end, but I could be making a mistake on any one of the steps above.  Or, I
> could be mis-interpreting what I should expect to get from the tls_as_xyz
> procedure.  I'd be happy to share details off-line with any one who might
> volunteer to help.

We have tools in Phenix to repeat steps "a" and "b", and I believe I can 
use tools in CCTBX constructor to put together some code to do "c". I 
think if I do it independently using one software frame-work to do all 
calculation steps that may help to see where things go out of sync..

All the best,

More information about the phenixbb mailing list