How many image duplications in a paper would be acceptable? If the paper has two identical photos that represent different experiments, and the authors’ reply is: ‘Oops, we uploaded the wrong photo’, that would be acceptable. Mistakes happen, and the authors can correct the error by sending in an erratum with the correct photo(s). But should editors be equally forgiving in the case of two cases of “Oops, we made a mistake”, or other, more complicated scenarios?
Let’s take a look at different types and numbers of duplications. As I explained in two previous blog posts (here and here), there are three categories of image duplications.
- Category I duplications: simple, identical duplications.
- Category II duplications: duplications involving shift, rotation, or a flip.
- Category III duplications: parts within the same panel are duplicated or parts from other panels are duplicated into another panel.
Category I is the most likely to be the result of an honest error, while Category III is really hard to explain by an honest error and the most likely to be done intentionally.
We can probably all agree that a paper with just one Category I duplication can be addressed with a simple correction or erratum. But how about papers with multiple duplications? Or, to sing along with Britney Spears, the “Oops! I did it again”, cases?
Let’s take a look at some different, real-life examples in order of increasing severity of error. Where would you draw the line as a journal editor presented with these concerns? Which paper could get away with a correction, and which paper should be retracted? There is no clear answer, so this post is meant as a discussion starter, not as a solution.
Two instances of type I duplications
Here is a figure with two duplications, marked in orange and pink. Note that the pink boxes represent different experiments but the same protein, while the orange boxes represent different experiment and different proteins. Should a paper with not just one, but two type I duplications be corrected, or retracted?
A single type II duplication
Here is a figure that has one type II duplication. The lanes marked in red appear to look very similar, suggesting that both actin panels are derived from the same blot, while the lanes represent different time points. Should a paper with this figure-of-concern be addressed with a correction or a retraction? Assume this is the only problem with this paper.Could this be an honest error or was this done with the intention-to-mislead? There is no clear answer here, so we might give the authors the benefit of the doubt and ask for a correction, if this was the only problem with a paper.
Two instances of type II duplications
Here is another set of duplications below. In this case, there are two sets of type II duplications within the same figure. Should a paper that contains this figure, with 2 pairs of shifted duplications, be corrected or retracted?
Combinations of type I and II duplications
In this example, a single figure had a type I duplication and a type II duplication within the same figure. The authors admitted they uploaded the wrong panels – twice. How would you respond to this? Would you consider both of these errors, one of which involved a 180 degrees rotation to be correctable, or do you think there might have been an intention to mislead? There is no clear answer here.
This paper has three figures that combined show two simple duplications (albeit of different exposure, marked in pink and pale purple) and multiple shifted duplications (marked in light and dark orange). If you are a journal editor, would you accept the excuse of an author that they made some mistakes while assembling or uploading the figure, or would you retract this paper?
https://pubpeer.com/publications/5887E2427572CFD0D9567BB16845D9
Type III duplications
In my opinion, a type III duplication should almost always lead to a retraction. Here are some examples. It is really hard to imagine that the duplications in the images below occurred in the Western blots, flow cytometry plot, or microscopy photo by accident.
Where would you draw the line?
There are lots of examples on PubPeer where the authors replied that they made a mistake. But after saying “Oops, another mistake” too many times, the other data in a paper become unreliable as well, and it becomes more and more likely that the mistakes were not done as the result of an honest error. But where does one draw the line? How many times can one say “Oops, I did it again”? How many mistakes can be addressed with a correction? At which point should an editor make the decision to retract the paper?
Again, I do not have a clear answer, but it is good to think about some threshold or flowchart.
Here is a proposal, meant to start the discussion, not to end it (edited):
- A single category I or II duplication might be addressed with a correction
- Two cases of a category I duplication within the same paper might be addressed with a correction
- Two category II duplications within the same paper could be a correction if there are many panels, or could be retracted
- Three or more category II duplications within the same paper should be retracted (‘Three strikes: You’re out’)
- Any combination of three or more category I/II duplications should be retracted
- Any paper with a category III duplication should be retracted
- If one group of authors has multiple papers with duplications, or if an institutional investigation has reveals misconduct in that group, it might make sense to be more strict, i.e. to retract papers even for one or two category II duplications.
Let me know what you think in the comments below.
Science Integrity Digest 
Nice post and I generally just wanted to say that I really appreciate all the good you are doing!
In terms of retract vs correct, I think there is also some room for responsiveness. I am not sure if I can imagine multiples of any of these errors happening innocently, but perhaps if lead/corresponding authors respond quickly with corrections even papers with I+II or II+II could avoid an automatic retraction. (quick response as evidence of non-guilty conscience?)
If academia is really publishing 2.5 million scientific papers per year then that means the clumsiest 0.01% still account for 250 papers annually, maybe plenty of room for accidental replication+rotation?
2.5 million source: http://blog.cdnsciencepub.com/21st-century-science-overload/ (lazy googling)
LikeLike
I agree with you that it may likely be impossible to draw a fair line. Each instance depends on technical issues of image preparation and the consequences it has for the main findings of the paper. On the other hand, a clear line may also help in making sure people pay more attention: we know from other areas how pernicious some incentives are we have in science. Why not use this to our advantage? A simple line of >1 and the paper is out, may lead to people quadruple-checking their work before submission and making sure fewer errors of sloppiness or haste are committed?
LikeLike
Interesting piece!
Personally I would like to see more liberal use of ‘flagged’ for category I and II. When the choice is between correct vs retract, the onus is on the journal to make a judgement call on the intent of the authors (malicious or not).
Saying that a paper has duplicate images is virtually undebatable and puts the editor on a much surer footing. This also lets the reader their own judgement about the data, and shifts the responsibility to the authors to do something about it.
Type III on the other hand….. Retract with minimal discussion! I simply can’t see any way in which these kind of images could occur without either malice or extreme incompetence, either of which should call the entire manuscript into question.
LikeLike
This book explains PI s don’t enter labs . See and hear Prof Jeffrey Hall’s Nobel Prize lecture. At the about 38th -40th minute in his lecture he explains about PI’s and what they do. He says not all PIs investigate
https://www.springer.com/in/book/9783030123253
The Rise of the Scientist-Bureaucrat
Survival Guide for Researchers in the 21st Century
Authors: Perez Velazquez, Jose Luis
LikeLike
Proc Natl Acad Sci U S A. 2001 Jul 31;98(16):9396-400.
Distinct mechanisms of cell cycle arrest control the decision between differentiation and senescence in human neuroblastoma cells.
Wainwright LJ1, Lasorella A, Iavarone A.
Author information
1
Department of Neurology, Developmental and Molecular Biology and Comprehensive Cancer Center, Albert Einstein College of Medicine, Bronx, NY 10461, USA.
Figure 4C.
Does this fit with Category III?
LikeLike
Proc Natl Acad Sci U S A. 2001 Jul 31;98(16):9396-400.
Figure 3B. Cdk6-bound cyclin D1 panel. Righmost 3 lanes much more similar than you would expect.
Would this fit Category III?
LikeLike
Genes Dev. 1995 Aug 1;9(15):1831-45.
Kip/Cip and Ink4 Cdk inhibitors cooperate to induce cell cycle arrest in response to TGF-beta.
Reynisdóttir I1, Polyak K, Iavarone A, Massagué J.
Author information
1
Cell Biology and Genetics Program, Memorial Sloan-Kettering Cancer Center, New York, New York 10021, USA.
Figure 1A.
Does this fit Category III?
LikeLike
J Biol Chem. 1997 Oct 10;272(41):25863-72.
Differential interaction of the cyclin-dependent kinase (Cdk) inhibitor p27Kip1 with cyclin A-Cdk2 and cyclin D2-Cdk4.
Blain SW1, Montalvo E, Massagué J.
Author information
1
Cell Biology Program and Howard Hughes Medical Institute, Memorial Sloan-Kettering Cancer Center, New York, New York 10021, USA.
Figure 1e.
Figure 2c.
Do these fit Category III?
LikeLike
Nucleic Acids Res. 2003 Sep 15;31(18):5377-88.
Biochemical evidence for Ku-independent backup pathways of NHEJ.
Wang H1, Perrault AR, Takeda Y, Qin W, Wang H, Iliakis G.
Author information
1
Department of Radiation Oncology, Division of Experimental Radiation Oncology, Kimmel Cancer Center, Jefferson Medical College, Philadelphia, PA 19107, USA.
Figure 4B.
Figure 4A.
Would these fit Category III?
LikeLike
I agree these figures do not look good but they do not appear to match the article you listed here. Could you please check?
LikeLike
You are quite right. The correct article is:
Nucleic Acids Res. 2004 Feb 3;32(2):767-75. Print 2004.
Involvement of Hus1 in the chain elongation step of DNA replication after exposure to camptothecin or ionizing radiation.
Wang X1, Guan J, Hu B, Weiss RS, Iliakis G, Wang Y.
Author information
1
Department of Radiation Oncology, Kimmel Cancer Center of Jefferson Medical College, Thomas Jefferson University, Philadelphia, PA 19107, USA.
Figure 4B.
Figure 4A.
LikeLike
Nucleic Acids Res. 2003 Sep 15;31(18):5377-88.
Biochemical evidence for Ku-independent backup pathways of NHEJ.
Wang H1, Perrault AR, Takeda Y, Qin W, Wang H, Iliakis G.
Author information
1
Department of Radiation Oncology, Division of Experimental Radiation Oncology, Kimmel Cancer Center, Jefferson Medical College, Philadelphia, PA 19107, USA.
Figure 2B.
Figure 3B.
LikeLike
Clin Cancer Res. 2010 Oct 1;16(19):4711-21. doi: 10.1158/1078-0432.CCR-10-0470. Epub 2010 Sep 21.Involvement of human herpesvirus-6 variant B in classic Hodgkin’s lymphoma via DR7 oncoprotein.Lacroix A1, Collot-Teixeira S, Mardivirin L, Jaccard A, Petit B, Piguet C, Sturtz F, Preux PM, Bordessoule D, Ranger-Rogez S.Author information1Centre Hospitalier Universitaire Dupuytren, Limoges, France.
Figure 5B. The bands and the areas close to the bands are very similar, yet the general background is different.
Would that fit Category III?
LikeLike
Cancer Res. 2002 Nov 1;62(21):6255-62.
Characterization of epithelial senescence by serial analysis of gene expression: identification of genes potentially involved in prostate cancer.
Untergasser G1, Koch HB, Menssen A, Hermeking H.
Author information
1
Molecular Oncology, Max-Planck-Institute of Biochemistry, Am Klopferspitz 18 A, D-82152 Martinsried/Munich, Germany.
Figure 1b.
Would the p53 panel fit Category III?
LikeLike
Cancer Res. 2005 May 15;65(10):4218-27.
Functional epigenomics identifies genes frequently silenced in prostate cancer.
Lodygin D1, Epanchintsev A, Menssen A, Diebold J, Hermeking H.
Author information
1
Molecular Oncology Group, Max-Planck-Institutes of Biochemistry, Martinsried and Institute of Pathology, Ludwig-Maximilians University, Munich, Germany.
Figure 1B.
Would this fit Category III?
Figure 1B again.
FYI earlier paper.
Oncogene. 2004 Dec 2;23(56):9034-41.
Prostate cancer is characterized by epigenetic silencing of 14-3-3sigma expression.
Lodygin D1, Diebold J, Hermeking H.
Author information
1
Molecular Oncology, Max-Planck-Institute of Biochemistry, Am Klopferspitz 18, D-82152 Martinsried/Munich, Germany.
LikeLike
I agree with these findings, but it would be more effective to post your concerns on PubPeer instead of commenting here.
LikeLike
Many thanks. I take your point.
The “Bik Scale for image duplication” may become a classic like the “Beaufort wind force scale”.
It helps to see the duplication(s) in proportion. I also take you point that there should be a proportionate response when image duplication is detected. This makes things fairer, and more likely that the general community of scientists, and people interested in science, will accept the results of any action taken to restore scientific integrity.
LikeLike
I don’t think Pubpeer is the right place. Some anoymous readers always ask the authors to comment the irregularities within the published figures. Pubpeer now cannot creat an annoynous account, so it is not fair for the authors to respond.
LikeLike
The 17 cases discussed in the CHM (Crystallography Horror Museum) may require other image duplication types. In front of these 17 obviously fake papers, the Nepalese authorities have said one month ago that a commission will give soon conclusions with media coverage. Up to now, nothing appeared but a retraction of the papers from ResearchGate… I was even suggested that I should write a paper in Nature and this would decide the commission to react faster ! This world is crazy.
LikeLike
And now a case of “Ooops we have changed the sense of the figure by replacing Q(A-1) by R(A)”. Paper not yet retracted : http://cristal.org/CHM/Mesli.pdf
LikeLike