This post was updated at 1145 UTC on August 13, 2021. In the original post, we noted that Joseph Powell and Gibran Hemani had not responded to our request for comment, which we sent shortly after learning under embargo from Nature that this retraction would be published. However, Powell did respond, copying Hemani, as Powell noted in a Twitter thread, and the email never reached us. We have added Powell’s comments, and updated the first sentence of the post to reflect them. We are also investigating why Powell’s email never arrived. We apologize for the errors regardless of the cause, and appreciate the opportunity to update.
The authors of a 2014 research letter in Nature have retracted their article, with near but not entire unanimity, after “new work led to interpretation of the original results being no longer fully valid,” according to the senior author.
Titled “Detection and replication of epistasis influencing transcription in humans,” the letter was written by a group from Australia, Europe and the United States led by Gibran Hemani, then of the University of Queensland, in Brisbane, and now of the University of Bristol, in the United Kingdom. The senior author on the paper was Joseph Powell, also then of Queensland but now at the Garvan Institute of Medical Research, in Darlinghurst, Australia. The paper has been cited 114 times, per Clarivate Analytics’ Web of Science.
According to the abstract of the article:
Here we show, using advanced computation and a gene expression study design, that many instances of epistasis are found between common single nucleotide polymorphisms (SNPs) … This study presents the first evidence, to our knowledge, for many instances of segregating common polymorphisms interacting to influence human traits.
As the retraction notice indicates, Nature in 2014 published a separate study that questioned the authors’ conclusions — a report that has been cited roughly half as often as the original article (63 times):
In this Letter, we reported the detection and replication of epistatic interactions between common single nucleotide polymorphisms (SNPs) that influence gene expression in peripheral blood, including both cis–cis and cis–trans interactions. We applied a statistical method widely used to detect epistasis. Wood et al. replicated these findings statistically in an additional whole-genome sequencing dataset but found that a large fraction of these epistatic effects could be explained by tagging sequence variants that were not genotyped in our study. They suggested that the interactions arose owing to haplotypes that tag single additive variants. In our response, we argued that such a mechanism could not explain cis–trans interactions. We have since undertaken further analyses to try to understand the mechanism that gives rise to cis–trans associations.
We find that in the presence of imperfectly tagged cis-expression quantitative trait loci with large additive effects, the F-test statistic used to detect interactions can result in an inflated false positive rate. As a result, we voice concern over whether our reported epistatic associations arose owing to biological mechanisms or from inflated test statistics caused by imperfectly tagged additive effects. All authors agree with the revised scientific interpretation of the original findings but opinions on whether to voluntarily retract the paper are mixed. Authors Gibran Hemani, Konstantin Shakhbazov, Harm-Jan Westra, Tonu Esko, Anjali K. Henders, Allan F. McRae, Jian Yang, Nicholas G. Martin, Andres Metspalu, Lude Franke, Grant W. Montgomery, Peter M. Visscher and Joseph E. Powell agree with the Retraction. Greg Gibson disagrees with the Retraction.
Senior author Powell told Retraction Watch:
The original paper was published in February 2014 and shortly afterwards we began some additional research into properties of the widely statistical test (F-test) that we had originally applied to identify epistatic interactions. We uncovered that under certain types of genetic architecture, the false-positive rate of the test can be larger than expected. We submitted this work to Nature in October 2015 as a proposed short Addendum as a means of alerting the research community to potential problems with this test. This submission, and subsequent updated versions and revisions, went through the peer review process. During this process we continued to conduct additional empirical and theoretical research to both more deeply understand the issues and try and identify solutions. This research was conducted between 2016 to 2020 and was also guided by feedback from anonymous peer reviewers. During this time the format of our submission changed to a Brief Communication Arising and then a Matters Arising as additional work was included. These subsequent formats were also peer reviewed. In consultation with the Editors at Nature, in November 2020 we decided that the best outcome would be a simultaneous voluntary author retraction of the original work, and publication of the new research as a Matter Arising paper.
Carefully understanding the problems associated with this test were challenging, and this underlies the period of time that the additional research was conducted over. The test we used in the original publication has widely been used to test for epistasis and had deep theoretical underpinning. To solve this problem required developing new theory, as well as substantial empirical research involving simulations under different types of genetic architecture. Because of the nature of the original research these simulation experiments require hundreds of billions of statistical tests, and took significant computational resources to conduct.
Some of the authors, myself and Gibran in particular, have moved institutions during this period. However, other than the typical impacts of relocating, these moves have not affected progress on this work. We have continued to work closely together on this problem and communicated frequently.
Gibson, a professor at Georgia Tech, forwarded us a letter he sent to Nature on July 14, 2020, detailing the reasons for his dissent. Gibson, the director of Georgia Tech’s Center for Integrative Genomics, asked us not to quote directly from the two-page document, in which he argued that retraction of the paper was not only unnecessary, but could harm scientific integrity and science more broadly by disincentivizing self-correction.
Using the threshold for retraction suggested by the move would mean as many as one in four papers in Nature and Science should be retracted, he wrote. Gibson also urged the editors to consider that retraction carries an unfair and lasting stigma and that Nature, by retracting the paper, would be indicting a peer review process that’s increasingly vulnerable to critique.
Like Retraction Watch? You can make a one-time tax-deductible contribution or a monthly tax-deductible donation to support our work, follow us on Twitter, like us on Facebook, add us to your RSS reader, or subscribe to our daily digest. If you find a retraction that’s not in our database, you can let us know here. For comments or feedback, email us at [email protected].
Peer Review has been increasingly vulnerable to critique because Peer Review has increasingly failed to do a competent job.
It’s not as though this paper is being retracted because of plagiarism or using bogus images or an author’s spouse was the reviewer. An error resulting from an artefact of the laboratory doesn’t cast aspersion on anyone.
The egregious reasons behind so many retractions is a shame but not a few of them should never have passed Peer Review.
Just an FYI, there’s no such thing as “Georgia Tech University.”
Fixed — thanks.