There are several issues under current discussion. We
outline a few of these below, in order of importance.
The structure 2hr0 is unambiguously fake. Valid arguments
have already been published in a Brief Communication by
Janssen et. al (Nature, 448:E1-E2, 9 August 2007). However,
the published response from the authors of the questioned
deposit may sound to unfamiliar person as an issue of a
scientific controversy. There are many additional
independent signs of intentional data fabrication in this
case, above and beyond those already mentioned.
One diagnostic is related to the fact that fabricating data
will not show proper features of proteins with respect to
disorder. The reported case has a very high ratio of
“Fobs”/atom parameters, thus the phase uncertainty is small.
In real structures fully solvent exposed chains without
stabilizing interactions display intrinsically high
disorder, yet in this structure these residues (e.g.,
Arg932B, Met1325B, Glu1138B, Arg459A, etc.) are impossibly
well ordered.
The second set of diagnostics is the observation of perfect
electron density around impossible geometries. For example,
the electron density is perfect (visible even at the 4 sigma
level in a 2Fo-Fc map) with no significant negative peaks in
an Fo-Fc map around the guanidinium group of Arg1112B, which
is in an outrageously close contact to carbon atoms of
Lys1117B. This observation appears in many other places in
the map as well. The issue is not the presence of bad
contacts, but the lack of disorder (high B-factors) or
negative peaks in an Fo-Fc map in this region that could
explain why the bad contacts remain in the model.
The third set of diagnostics are statistics that do not
occur in real structures. The ones mentioned previously are
already very convincing (moments, B-factor plots, bulk
solvent issues, etc.). We can add more evidence from a round
of Refmac refinement of the deposited model versus the
deposited structure factors. The anisotropic scaling factor
obtained, which for a structure in a low symmetry space
group such as C2 that has an inherent lack of constraint in
packing symmetry, is unreasonable (particularly in view of
the problems with lattice contacts already mentioned). The
values from a Refmac refinement for a typical structure in
space group C2 are: B11 = 0.72 B22 = 1.15 B33 = -2.12 B12
= 0.00 B13 = -1.40 B23 = 0.00 (B12 and B23 are zero due to
C2 space group symmetry). For structure 2hr0: B11 = -0.02
B22 = 0.00 B33 = 0.02 B12 = 0.00 B13 = 0.01 B23 = 0.00.
Statistical reasoning can lead to P-values in the range of
10exp(-6) for such values to be produced by chance in a real
structure, but they are highly likely in a fabricated case.
The fourth set of diagnostics are significant
inconsistencies in published methods, e.g. the authors claim
that they collected data from four crystals, yet their data
merging statistics show an R-merge = 0.11 in the last
resolution shell. It is simply impossible to get such values
particularly when I/sigma(I) for the last resolution shell
was stated as 1.32. Moreover, the overall I/sigma(I) for all
data is 5.36 and the overall R-merge is 0.07 – values highly
inconsistent with the reported data resolution, quality of
map and high data completeness (97.3%).
Overall this is just a short list of problems, the
indicators of data fabrication/falsification are plentiful
and if needed can be easily provided to interested parties.
We fully support Randy Read's excellent comments with our
view of retraction and public discussion of this problem:
“Originally I expected that the publication of our Brief
Communication in Nature would stimulate a lot of discussion
on the bulletin board, but clearly it hasn't. One reason is
probably that we couldn't be as forthright as we wished to
be. For its own good reasons, Nature did not allow us to use
the word "fabricated". Nor were we allowed to discuss other
structures from the same group, if they weren't published in
Nature.”
One needs to address this policy with publishers in cases of
intentional fraud that can be proven simply by an analysis
of the published results. At this point the article needs to
be retracted by Nature after Nature's internal investigation
with input from crystallographic community rather then after
obtaining results of any potential administrative
investigation of fraud.
“Another reason is an understandable reluctance to make
allegations in public, and the CCP4 bulletin board probably
isn't the best place to do that.”
The discussion of fraud allegation was initiated by public
reply to a question addressed to a single person, so it
happened by chance rather than by intention, but with no
complaint from our side.
On a different aspect of the discussion – namely, data
preservation—currently, funding agencies as well as
scientific responsibility requires authors of any
publication to preserve and keep for a reasonable amount of
time all the primary data for experiments. In the case of
crystallography, this would reasonably be interpreted to
mean preservation of X-ray diffraction images. It should be
required that authors retrieve and provide these data in
case of controversy. In any case, the key results are
deposited in the PDB. As long as structure factors are
deposited it is effectively impossible to do a credible job
of fabricating data, as has been discussed.
G. Sheldrick:” The deposited structure 2HR0 shows all the
signs of having been refined, deliberately or accidentally,
against 'calculated' data. The model used to 'calculate' the
data had (almost) constant B values in a rather empty
cell containing no solvent. For example, it could have been
a (partial?) molecular replacement solution obtained using
real data. It seems to me that it is perfectly possible that
two reflection files (or two columns in an mtz file) were
carelessly exchanged by a crystallographically inexperienced
researcher.”
This scenario is impossible. If this had happened, then
subsequent refinement (which we did) of the structure would
produce impossibly low R-factors. The calculated structure
factors were randomized to prevent this occurrence,
indicating strongly that it was done intentionally. There
are more indicators of intentional fabrication/falsification
of data. The structure factors used in refinement are
clearly derived from a model of low quality, which suggest a
model derived from a low resolution dataset. But the most
convincing indicator is the form of the response to the
Brief Communication by Janssen et.al. If it would be any
kind of mistake it would have to be of a very obvious
nature, and the authors would recognize it and state it in a
public response. Their defense does not make any sense in
terms of crystallography when combined with inspection of
the deposited data.
Dominika Borek
UT Southwestern Medical Center
5323 Harry Hines Boulevard
Dallas, Texas 75390-8816
phone: +1 214-645-6378
fax: +1 214-645-6353
[EMAIL PROTECTED]