Hi,
I am getting a "non-positive definite" error when following the mrc refmac5 Tutorial:
https://www2.mrc-lmb.cam.ac.uk/groups/murshudov/content/tutorials/refmac_tutorial/files/part_2.html#prepare_tls
I'm using the same data recommended by the tutorial and am following the tutorial with no deviations -see error below. Am I misinterpreting the SKTTLS summary, is this a binary problem, a data problem, tutorial problem... Any guidance would be appreciated.
Best wishes,
Reza
#### SKTTLS SUMMARY ####
# Total number of bonds between residues: 491
# Number of bonds beyond 95th percentile for any residual: 0
# Number of bonds beyond 99th percentile for any residual: 0
# NOTE: bonds beyond 99th %ile are included in the 95th %ile count
# Percentiles are from REFMAC segmented TLS refinements in PDB as of Sept. 2009
# Residuals are calculated for each bond between residues; one additional
# residual is calculated for each alternate atom in such bonds, from the first
# C or O3' to the alternate atom.
# Extreme residuals may indicate problems with the ADPs for the two atoms in the bond:
# -- ADPs differ by more than usual, which may be due to misassignment of TLS
# segment boundaries, or
# -- something went wrong during refinement - maybe the residual B bottomed out
# at 2.0? - or
# -- one or both atoms have non-positive definite ADP, in which case CCuij is
# given the arbitrary extreme value of zero - regardless of how
# similar the two ADPs are.
# Statistics for residuals in this structure:
#
# RESIDUAL MEAN SIGMA WORST
# CCuij 0.9993 0.0009 0.9945
# RSIMU 0.0313 0.0211 0.1235
# RDELU 0.0305 0.0255 0.1402
# Statistics for anisotropy (ratios of smallest to largest eigenvalue)
# for anisotropic positive-definite atoms in this structure:
#
# NUMBER
# CHAIN TYPE MEAN SIGMA OF ATOMS
# Protein 0.6586 0.0830 3950
# Nucleic Acid 0.0000 0.0000 0
# Other nonsolvent 0.0000 0.0000 0
#### END OF SKTTLS SUMMARY ####
Reza Khayat, PhD
Assistant Professor
City College of New York
Department of Chemistry
New York, NY 10031
________________________________________
From: CCP4 bulletin board <[log in to unmask]> on behalf of Dmytro Guzenko <[log in to unmask]>
Sent: Monday, April 10, 2017 6:29 PM
To: [log in to unmask]
Subject: Re: [ccp4bb] Fwd: Re: [ccp4bb] Structure comparison
Gert,
There is nothing wrong with suggesting 45 softwares, as one of those may
do exactly what a person needs and save loads of time.
Besides, comparing protein structures locally need not involve any
superposition method. I would say comparing torsion angles/distances
between atoms is a more common solution to this problem. There are still
cutoffs and weights to consider of course. Approaches vary on how to
handle those (e.g. theoretical or through validation), but I seriously
doubt your response constitutes 'case closed'.
Kind regards,
Dmytro.
On 10/04/17 12:06, Gert Vriend wrote:
> Rather then now start mentioning yet 45 other 3D superposition
> softwares, I think a pointer to the Wikipedia should suffice:
>
> https://en.wikipedia.org/wiki/Structural_alignment
>
> The Russian Doll effect (if you align more residues the comparison stats
> get worse) is best explained (my opinion) by the CATH people (Orengo
> group), and Arthur Lest is the only one (still, I think) who has been
> thinking extensively about the cutoff question.
>
> Greetings
> Gert
>
>>>> From: Reza Khayat
>>>> Sent: Sunday, April 9, 2017 6:07 PM
>>>> To: CCP4 bulletin board
>>>> Subject: Structure comparison
>>>>
>>>> Hi,
>>>>
>>>> I have refined several structures of a protein from different space
>>>> groups and would like to compare them to one another. Is there a
>>>> program/software suite that would provide an objective comparison
>>>> of the structures and identify regions where the structures are
>>>> sufficiently different from one another to warrant a closer look? I
>>>> think the most important aspect of the analysis would be defining a
>>>> threshold (possibly based on resolution and structure statistics)
>>>> that would identify sufficient difference between structures. Thanks.
>>>>
>>>> Best wishes,
>>>> Reza
>>>>
>>>> Reza Khayat, PhD
>>>> Assistant Professor
>>>> City College of New York
>>>> Department of Chemistry
>>>> New York, NY 10031
>
> Het Radboudumc staat geregistreerd bij de Kamer van Koophandel in het
> handelsregister onder nummer 41055629.
> The Radboud university medical center is listed in the Commercial
> Register of the Chamber of Commerce under file number 41055629.
|