Tackling paper mills requires us to prevent future contamination and clean up the past – the case of the journal Bioengineered

ABSTRACT

Taylor & Francis journal Bioengineered has been targeted by paper mills. Our goal is to identify problematic articles published in Bioengineered during the period 2010 to 2024. Dimensions was used to search for articles that contained the terms ‘mouse’ OR ‘mice’ OR ‘rat’ OR ‘rats’ in title or abstract, published in Bioengineered between January 1st 2010 to December 31st 2024. All articles were assessed by eye and by using software to detect inappropriate image duplication and manipulation. An article was classified as problematic if it contained inappropriate image duplication or manipulation or had been previously retracted. Problematic articles were reported on PubPeer by the authors if they had not been reported previously. All included articles were assessed for post-publication editorial decisions. We have excluded all articles published in 2024 from further analysis, as these were all retraction notices. We assessed the remaining 878 articles, of which 226 (25.7%) were identified as problematic, of which 35 had been previously retracted. One retracted article was later de-retracted. One article received a correction. None of the included articles received an expression of concern or the Taylor & Francis ‘under investigation’ pop-up. Taylor & Francis’ lack of visible editorial action has left the scientific community vulnerable to reading and citing hundreds of problematic articles published in Bioengineered. To uphold scientific integrity, Taylor & Francis should use the findings of this study as a starting point to systematically identify all compromised articles in Bioengineered and take appropriate editorial action.

KEY POINTS

• Taylor & Francis noted that their journal Bioengineered was targeted by paper mills.

• All articles published in Bioengineered between January 1st 2010 to December 31st 2023 containing the terms “mouse” OR “mice” OR “rat” OR “rats” in title or abstract were assessed for inappropriate image duplication and manipulation using ImageTwin and visual inspection.

• Among the 878 included articles, 226 (25.7%) were classified as problematic.

• Actions taken by the publisher appear to have stemmed the tide of new paper mill submissions, but a backlog of contaminated articles remains in the literature.

• Taylor & Francis’ lack of visible editorial action has left the scientific community vulnerable to reading and citing hundreds of problematic articles published in Bioengineered.

Introduction

Paper mills are manipulating the scientific record by selling authorship of poor-quality or fabricated manuscripts [1]. Services provided by paper mills may include the sale of manuscripts, citation schemes, fake peer review, and data fabrication [2–4]. Because paper mill manuscripts often are based on fabricated data, they can produce high numbers of articles in a short time span. When paper mills target a scientific journal, it can lead to a spectacular increase in the number of published articles in that journal [5]. An example of such an increase can be seen in the Taylor & Francis journal Bioengineered.

Bioengineered saw a ten-fold increase in the number of published articles in 2021 and 2022 compared to the previous years (see Figure 1). Taylor & Francis acknowledged in a blog post (published somewhere in 2023 by an unknown author, unreferenced) that Bioengineered was targeted by paper mills and stopped publishing paper mill produced articles from the start of 2023.

Figure 1.

Number of published articles in Bioengineered from its inception in 2010 until the end of 2024. Source: Dimension.Ai.

Although Taylor & Francis has seemingly been successful in stopping the publication of large amounts of new paper mill articles, they have not retracted all the problematic articles that have been published in the years 2021 and 2022. This is puzzling, as they have clearly stated in their blog post (published somewhere in 2023 by an unknown author, unreferenced) that they noticed issues in Bioengineered since early 2021, which is now 4 years ago.

The goal of this study was to identify problematic articles published in Bioengineered during the period 2010–2024. In our search, we focused on articles with photographic images, because duplications in photos within or across papers are objective indicators of sloppiness or an intention to mislead. Specifically, duplicated or overlapping photos found in multiple papers from different groups of authors at different institutions point toward a common source selling data or complete papers, a feature associated with paper mill activity [4]. We have also assessed how often an editorial decision has been made to correct or retract problematic articles that were published during these years (2010–2024).

Methods

Search

We performed several searches in Dimensions (Dimensions.ai, Digital Science, London, UK) on December 16 2024, March 19 2025 and March 20 2025. Our searches included articles published in the journal Bioengineered between 1 January 2010 to 31 December 2024, that contained the terms ‘mouse’ OR ‘mice’ OR ’rat’ OR ‘rats’ in title or abstract. We exported the results of our searches as Excel files (Microsoft, Redmond, USA) via the Dimensions.ai interface, which we later merged into one master Excel file. The decision to focus on articles containing these terms was made because they frequently include figures such as histology images and Western immunoblots, which are suitable for assessing inappropriate image duplication or manipulation. Retraction notices were excluded from the analysis.

Identification of problematic articles

All authors independently evaluated the included studies using ImageTwin (ImageTwin AI GmbH, Vienna, Austria) and visual inspection between 16 December 2024 and 23 March 2025. ImageTwin is a software tool designed to detect (partial) overlaps within and between figures in scientific articles. It also identifies (partially) overlapping figures across different articles using its database of over 75 million scientific images [6]. In order to keep the number of false-positive findings to a minimum, all image-related issues flagged by ImageTwin had to be visually confirmed by at least one of the authors. When in doubt about a finding, it was discussed with at least one other author. When still in doubt after discussion, the issue would not be classified as problematic.

An article was classified as problematic if it contained inappropriate image duplication or manipulation or had been previously retracted.

Reporting of problematic articles

Articles with identified image-related issues were reported on PubPeer (Pubpeer.com, The PubPeer Foundation, California, USA) [7] unless they had already been flagged by others. Additionally, we documented other types of issues, such as plagiarism (See this post, 8 November 2019 posting by Elisabeth M. Bik to the Science Integrity Digest blog, unreferenced), tortured phrases [8], mismatched primers [9], incorrect methods (drawing from our own expertise) on PubPeer when encountered. However, articles with only these non-image-related issues (and that were not previously retracted) were not classified as problematic for the purposes of this study.

To ensure completeness, one author (RA) reevaluated all included articles on 23 March 2025, to check for any new reports on PubPeer.

Classification of image problems

Image-related issues were classified into three categories:

• Image duplication within the same figure (either within or between panels of the same figure).

• Image duplication between figures within the same article.

• Image duplication between different articles.

Editorial decisions

One author (RA) accessed all included articles through the Taylor & Francis website on March 23 and 24 March 2025 to assess if any article had been retracted or corrected, or if any article received an expression of concern or the Taylor & Francis ‘under investigation’ pop-up [10].

Outcome measures

We analyzed three outcome measures:

1. The number of included articles relative to the total number of articles published in Bioengineered, categorized per year.

2. The proportion of included articles identified as containing problematic images, categorized per year.

3. Any editorial decisions made for included articles (e.g. correction, expression of concern, or retraction).

Results

Included articles

Our search identified a total of 885 articles. After the removal of 7 retraction notices, all published in 2024, 878 articles remained for further assessment. Because the 7 excluded retraction notices comprised all the articles from 2024 identified by our search, we decided to exclude 2024 from our analysis in its entirety (Table 1 and Supplementary file 1).

Table 1.

Number of articles assessed and number of problematic articles identified.

Year	Articles published – n	Articles included in our sample – n (%)	Problematic articles included in our sample – n (%)
2010	67	10 (14.9%)	1 (10.0%)
2011	66	7 (10.6%)	0 (0.0%)
2012	69	8 (8.3%)	0 (0.0%)
2013	94	9 (9.6%)	1 (11.1%)
2014	60	6 (10.0%)	0 (0.0%)
2015	64	3 (4.7%)	0 (0.0%)
2016	119	11 (9.2%)	1 (9.1%)
2017	75	1 (1.3%)	0 (0.0%)
2018	6	0 (0.0%)	N/A
2019	70	14 (20.0%)	5 (35.7%)
2020	131	22 (16.8%)	11 (50.0%)
2021	1072	391 (36.5%)	101 (25.8%)
2022	1072	394 (36.8%)	105 (26.6%)
2023	64	2 (3.1%)	1 (50.0%)
TOTAL	3029	878	226

Problematic articles

Among the 878 included articles, 226 (25.7%) were identified as problematic (see Table 1, Figure 2 and Supplementary file 1). Of these, 194 contained image-related issues and 35 had been previously retracted (not all retracted articles contained image-related issues).

Figure 2.

Included Bioengineered articles in our sample, divided in problematic articles (red, dashed bars) and non-problematic articles (green, solid bars).

The 194 articles with image-related issues were categorized as follows:

• Image duplication within the same figure − 78 articles;

• Image duplication between figures within the same article − 32 articles;

• Image duplication between different articles − 122 articles.

Some articles contain several types of image-related issues, which is indicated in Supplementary file 1.

Number of editorial decisions taken by Bioengineered

Of all the included articles, a total of 35 articles were retracted. One article (doi: 10.1080/21655979.2021.1987083) had previously been retracted in January 2024 but was ‘de-retracted’ in March 2024 (Supplementary File 1). One article had been previously corrected (doi: 10.1080/21655979.2016.1238534). For both these articles additional issues were found and both should be further investigated by Taylor & Francis.

None of the included articles received an expression of concern or the Taylor & Francis ‘under investigation’ pop-up [10].

Discussion

Our study identified a large number of problematic articles (226 out of 878, 25.7%) in the Taylor & Francis journal Bioengineered.

Shift in publication trends

It seems as if Bioengineered was already targeted by paper mills before 2021. In the years prior to 2019, the number of articles containing the terms ‘mouse,’ ‘mice,’ ‘rat’ or ‘rats’ in title or abstract always remained under 15% of the total number of published articles. However, in 2019 and 2020 this increased to over 15% and in 2021 and 2022 the numbers even surpassed 35%. We hypothesize that paper mills are specifically publishing articles resembling preclinical animal studies, possibly to meet the requirements of their customers. This hypothesis needs further verification in future research.

Issues preceding 2021

Our results suggest that paper mills -or other authors producing sloppy or potentially fraudulent research- had already targeted Bioengineered before the surge in accepted articles in 2021 and 2022. The relatively stable number of publications in 2010–2020 suggests that paper mills may have been ‘testing the waters’ to gauge what could pass peer review. We have postulated several hypotheses to explain the sudden increase in the number of published articles:

• Possibly, multiple additional paper mills found their way to Bioengineered early 2021;

• Maybe paper mills were waiting for editors to recognize that many authors were eager to publish in Bioengineered, leading to editorial policy changes, which allowed for steep increases in the number of published articles;

• Perhaps paper mills were trying to influence editors with financial rewards to manipulate the manuscript acceptance process, which has been described previously [11].

These hypotheses will probably remain untested until Taylor & Francis investigates these themselves or if they choose to release the relevant data to the scientific community.

Underestimated scope of the problem

Our study provides a conservative estimate of the actual number of problematic articles in Bioengineered. Beyond the 226 flagged articles, we identified an additional 67 flagged articles with other serious issues, such as tortured phrases, mismatched primers, and incorrect methods.

Furthermore, a troubling pattern emerged: many articles appeared questionable even in the absence of inappropriate image duplication or manipulation. For example, many articles contained Western immunoblots that apparently did not contain any visible duplications, but that looked unrealistic. These immunoblots often contained band shapes resembling those observed in the ‘Tadpole paper mill’ [5], but without the repetitive backgrounds. Such papers were not listed in our analysis as problematic due to lack of duplications or other objective problems. However, the unnatural pattern of bands suggested they could have been digitally generated using algorithmic methods.

Similarities in geography and title structures

We were struck by the prevalence of author affiliations with (regional) hospitals and universities in China, raising the question why there was such a lack of geographic diversity. Additionally, article titles and image layouts often followed highly similar templates, with titles commonly structured as: [Non-coding RNA] + [simple present tense] + [disease] + [by/through/via] + [present continuous tense] + [pathway/process].

Some examples include:

• LncRNA SNHG1 promotes tumor progression and cisplatin resistance through epigenetically silencing miR-381 in breast cancer

• MiR-211 protects cerebral ischemia/reperfusion injury by inhibiting cell apoptosis

• Downregulated circular RNA hsa_circ_0005797 inhibits endometrial cancer by modulating microRNA-298/Catenin delta 1 signaling

• Circ_0017639 facilitates proliferative, migratory, and invasive potential of non-small cell lung cancer (NSCLC) cells via PI3K/AKT signaling pathway

• Human umbilical cord-mesenchymal stem cells-derived exosomes carrying microRNA-15a-5p possess therapeutic effects on Wilms tumor via regulating septin 2

The absence of geographical diversity, combined with the formulaic writing style, and recurring image layouts, strongly suggests coordinated efforts and template-based writing – likely by paper mills [1].

Taylor & Francis has not fully addressed the paper mill problem

In a blog post (published somewhere in 2023 by an unknown author, unreferenced) on the Taylor & Francis website, Todd Hummel (Taylor & Francis Global Publishing Director, STM) stated that Bioengineered had ‘overcome the paper mill problem.’ However, their actions make it clear that they have been wholly focused on preventing publication of new articles from paper mills, by methods such as integrity checks at submission and vetting of reviewers. While we acknowledge the publisher’s effort to improve the submission, peer-review, and publication process, our findings indicate that they have not sufficiently addressed problematic articles already published.

One reason may be because it is much harder to retract an already-published paper than to trap paper mill articles at the point of submission. Journal staff need the expertise to recognize when an image problem is highly unlikely to be due to ‘honest error.’ They need to communicate with authors regarding retraction and may encounter authors who are litigious or non-responsive. We understand that large-scale, journal-wide investigations are complex, time-consuming, and difficult for all parties involved. However, it remains unclear whether Taylor & Francis is actively investigating these articles. None of the flagged articles display the “under investigation” pop-up, which could serve as a simple yet effective way to alert readers and researchers [10]. Given that Taylor & Francis generates hundreds of millions of pounds in annual revenue [12], the publisher has both the resources and the responsibility to systematically investigate Bioengineered´s archives and retract compromised articles.

There are two reasons why it is important to retract published paper mill articles. First, these articles destroy the integrity of the publication record. They may get cited and find their way into systematic reviews, subverting attempts to integrate the literature. Second, retraction of published articles undermines the business model of paper mills. If it happens often enough, paying customers may be harder to find, because they will hear of others who have paid to have an article published, only to have it later retracted. The customer has no means of getting their money back, and instead of the benefit of a journal publication, they find themselves associated with the stigma of enforced retraction.

Consequences of Bioengineered´s decline

The issues at Bioengineered have had significant repercussions. The journal is now classified as category 0 (not approved) by the Norwegian Register for Scientific Journals [13]. In addition, the journal was listed on the early warning journal list [14] due to paper mill activity, by the National Science Library of the Chinese Academy of Sciences in February 2024. Bioengineered has also been delisted from Web of Science’s Master Journal List in April 2025 after being ‘on hold’ since March 2024. During this 12-month period, Taylor & Francis had the opportunity to solve Bioengineered’s problems but it seems they were unable to prevent the delisting. This resulted in a loss of Bioengineered’s impact factor as well as losing an important way for researchers and the public to access the journal’s content.

Another major concern is that nearly all of the problematic articles continue to be cited. Only 3 of the 226 flagged articles have never been cited. The remaining 223 were cited between 1 and 117 times (Supplementary File 1). Many researchers may be unaware of what has transpired at Bioengineered, meaning these problematic articles are still being referenced in scientific literature. Additional research is needed to investigate the effects of citing these problematic articles.

Limitations

Our study has several limitations:

1). Sampling bias

First, we have only investigated a subset of Bioengineered’s publications, specifically articles related to rodent studies. Paper mills may have targeted these types of studies more heavily, but other disciplines within the journal may also be affected. Our decision to focus on rodent-related articles means other problematic articles may remain undetected. Alternatively, we could have used a full text search of terms such as ‘immunohistochemistry’ or ‘western blot’ in Dimensions, which might have increased the study pool. It is unclear if this would have a meaningful effect on the fraction of problematic articles.

2). Potential false positives and false negatives

We took great care to minimize false positives by focusing on image manipulations that are relatively easy for readers to verify (see PubPeer links in Supplementary File 1). However, false negatives are possible – some problematic articles may have escaped detection due to more subtle manipulation techniques. Many articles had highly similar figure styles, suggesting common authorship or outsourced manuscript preparation, but not all contained explicit evidence of image duplication.

3). Paper mill attribution

In some cases, we have strong evidence of paper mill involvement – for example different authors using the same figure in multiple articles. However, we do not have a clear connection to paper mills for all problematic articles. Thus, it is possible that some articles come from a different type of source.

The path ahead

After publishing the manuscript on a pre-print server, we submitted it to Bioengineered to give Taylor & Francis a chance to inform their readers about the issues that we have identified in Bioengineered. Our work serves as an example of how the work of scientists can be used as input for publishers to improve the scientific record. As previously stated, ‘publishers, journals, researchers, and institutions must work together and show courage and determination to recognize and ultimately reject the worthless magic of papers created without experiments’ [5].

Conclusions

Taylor & Francis’ lack of visible editorial action has left the scientific community vulnerable to reading and citing hundreds of problematic articles published in Bioengineered. To uphold scientific integrity, Taylor & Francis should use the findings of this study as a starting point to systematically identify all compromised articles in Bioengineered and take appropriate editorial action.

Disclosure statement

All authors have ongoing collaborations with ImageTwin and have been given free access in return. Both David Bimler and Elisabeth M. Bik receive donations through Patreon to encourage their sleuthing activities. Elisabeth M. Bik receives speaker fees and travel reimbursement to give talks and workshops and does occasional consulting work for research institutions, funders, and publishers.

Data availability statement

The authors confirm that the data supporting the findings of this study are available within the supplementary materials. Data are also available on Zenodo: https://doi.org/10.5281/zenodo.15671465.

Author contributions

RA: conceptualization, data curation, formal analysis, investigation, methodology, supervision, validation, writing – original draft, writing – review and editing.

EMB: conceptualization, data curation, investigation, validation, writing – original draft, writing – review and editing.

DB: conceptualization, data curation, investigation, validation, writing – original draft, writing – review and editing.

MPO: conceptualization, data curation, investigation, validation, writing – original draft, writing – review and editing.

KP: conceptualization, data curation, investigation, validation, writing – original draft, writing – review and editing.

Preprint statement

This manuscript has been published as a preprint on arXiv. DOI: https://10.48550/arXiv.2503.21267.