Constructive correction

Biology research generates complex data that often provide at best narrow glimpses into immensely complicated systems—be it at the level of molecules, cells, organisms, or ecosystems. It is not just the known unknowns, but the unknown unknowns that complicate mechanistic interpretation. Before regulatory ncRNAs and splice variants, we simply blissfully ignored the additional complexity to protein regulatory cascades. It is therefore to be expected that even cautious interpretation of limited datasets is subject to later refinement and revision. This is what makes biology research so exciting and fulfilling, and what makes it so much more effective as a community effort.

It also necessitates a scientific process that allows for – and indeed encourages – iterative refinement and correction. Such correction can be re‐interpretation of a given dataset or a change in understanding based on broader, deeper, or orthogonal data. There is no room for “dogma” in a healthy scientific ecosystem since the data available at any one time provide insight that is typically too limited for absolutist conclusions. As a community, we need to be open to adapt our ideas and thinking, and thus our papers. This is asking a lot since scientific ideas are linked to the success of individuals and prestige.

Fiona Watt covers a number of related points in her commentary How to avoid mistakes in science in this issue (Watt, 2023). After discussing how to prevent mistakes pre‐publication and how to deal with them post‐publication, Watt concludes that “mistakes happen, but science is self‐correcting.” This is indeed a fundamental tenet of the scientific process and the community needs to ensure that it remains true. Self‐correction by authors should be as low‐friction as possible and the self‐correction of “honest mistakes” should be rewarded in research assessment evaluations. Instead, Watt notes that “scientists may be slow to own up to a mistake because they fear it will cause reputational harm.” Indeed, there is a perception of little to gain and much to lose when self‐correcting. Still, once one's reputation is at stake, many a senior author will be compelled to act. For early career researchers, however, a retraction or even substantive correction will be perceived as a career‐altering event.

What can journals do to support both self‐correction by authors and by the community? We have discussed this matter for a decade at EMBO Press, culminating in a workshop that recommended to make an explicit distinction between an author initiated‐ and an externally initiated‐ or even forced retraction. We use the term “withdrawal” for the fomer and reserve “retraction” for the latter. We have lowered the bar to self‐correction further by encouraging withdrawal/retraction of only the affected data and the linked descriptions and conclusions, as long as it does not undermine overall confidence in the body of work. In cases where correct data did exist at the time of acceptance, we encourage replacing faulty with correct data (Wilfong Boxheimer & Pulverer, 2019). These measures alone are unlikely to change self‐correction culture though. They need to be matched by the research community explicitly distinguishing between bona fide research‐integrity breaches and “honest mistakes” or correction following reinterpretation of data. Watt discusses the lab support necessary to help junior scientists through this moral minefield.

At the higher level of community‐based self‐correction, we support the commenting platform PubPeer and work through all substantive issues in accordance with our research integrity and corrections policies. In discussing how post‐publication commenting can help identify errors that need to be corrected, Watt argues “some journals still cling to the concept of a paper being an end‐stage scholarly record as opposed to being updatable.” Scientific publishing is undoubtedly still beholden to concepts from the cellulose‐based article of the last century. Change is needed, but I suggest to nuance this recommendation by noting that dissemination of information focused around the “research paper” does continue to provide a clearly traceable, archival “scholarly record,” albeit with a visible change history (versioning). To achieve this goal, EMBO Press articles have become more dynamic by allowing traceable in‐line updates to mistakes and errors, ranging from self‐contained but consequential issues such as the archetypical typo of using milli‐ instead of micro or misspelt names, to larger‐scale issues such as misidentification of a pseudogene as a coding gene (Gal et al, 2017) or figure panels with unreliable data (Bechara et al, 2023). In such cases, changes in text and figures are marked in the article for every reader to see. Admittedly, this is not quite fully‐fledged versioning and is associated with a separately published “correction”.

If versioning at the article level works, why are we still posting clunky “corrections”, which are hard to follow by even the most ardent reader. The reason is that the research article is connected to other scholarly databases and platforms, such as PubMed, PubMedCentral, or ResearchGate. These platforms will not be updated without the trigger of a formal correction.

Importantly, full versioning is available not only at preprint level, but also for the database repository versions of EMBO Press papers which capture the source data underlying figures in a research paper on BioStudies. Importantly, as mentioned by Watt, this mechanism also allows attribution of specific experiments, crucial for both credit and accountability. Furthermore, EMBO Press is part of PRO‐MAP, which has called for methods and protocols sections to be versioned (preprint: Leite et al, 2023). As Watt stated, “the requirement for data availability sections in journals and reproducible methods and protocols both help mitigate mistakes and facilitate the reproducibility of published experiments” – EMBO Press requires just that.

EMBO's scientific publications are adapting to better address the key elements of the self‐correction paradigm of science highlighted by Watt. I believe we have come some way since Watt's experience of lecturers deleting figures in a single printed copy of their paper with a stroke of a fountain pen.