With the growth of artificial intelligence (AI)-ready datasets such as the National Health and Nutrition Examination Survey (NHANES), new opportunities for data-driven research are being created, but also generating risks of data exploitation by paper mills. In this work, we focus on two areas of potential concern for AI-supported research efforts. First, we describe the production of large numbers of formulaic single-factor analyses, relating single predictors to specific health conditions, where multifactorial approaches would be more appropriate. Employing AI-supported single-factor approaches removes context from research, fails to capture interactions, avoids false discovery correction, and is an approach that can easily be adopted by paper mills. Second, we identify risks of selective data usage, such as analyzing limited date ranges or cohort subsets without clear justification, suggestive of data dredging, and post-hoc hypothesis formation. Using a systematic literature search for single-factor analyses, we identified 341 NHANES-derived research papers published over the past decade, each proposing an association between a predictor and a health condition from the wide range contained within NHANES. We found evidence that research failed to take account of multifactorial relationships, that manuscripts did not account for the risks of false discoveries, and that researchers selectively extracted data from NHANES rather than utilizing the full range of data available. Given the explosion of AI-assisted productivity in published manuscripts (the systematic search strategy used here identified an average of 4 papers per annum from 2014 to 2021, but 190 in 2024–9 October alone), we highlight a set of best practices to address these concerns, aimed at researchers, data controllers, publishers, and peer reviewers, to encourage improved statistical practices and mitigate the risks of paper mills using AI-assisted workflows to introduce low-quality manuscripts to the scientific literature.

The effect of COVID-19 infection on the human metabolome has been widely reported, but to date all such studies have focused on a single wave of infection. COVID-19 has generated numerous waves of disease with different clinical presentations, and therefore it is pertinent to explore whether metabolic disturbance changes accordingly, to gain a better understanding of its impact on host metabolism and enable better treatments. This work used a targeted metabolomics platform (Biocrates Life Sciences) to analyze the serum of 164 hospitalized patients, 123 with confirmed positive COVID-19 RT-PCR tests and 41 providing negative tests, across two waves of infection. Seven COVID-19-positive patients also provided longitudinal samples 2–7 months after infection. Changes to metabolites and lipids between positive and negative patients were found to be dependent on collection wave. A machine learning model identified six metabolites that were robust in diagnosing positive patients across both waves of infection: TG (22:1_32:5), TG (18:0_36:3), glutamic acid (Glu), glycolithocholic acid (GLCA), aspartic acid (Asp) and methionine sulfoxide (Met-SO), with an accuracy of 91%. Although some metabolites (TG (18:0_36:3) and Asp) returned to normal after infection, glutamic acid was still dysregulated in the longitudinal samples. This work demonstrates, for the first time, that metabolic dysregulation has partially changed over the course of the pandemic, reflecting changes in variants, clinical presentation and treatment regimes. It also shows that some metabolic changes are robust across waves, and these can differentiate COVID-19-positive individuals from controls in a hospital setting. This research also supports the hypothesis that some metabolic pathways are disrupted several months after COVID-19 infection.

The COVID-19 pandemic has led to an unprecedented demand for testing - for diagnosis and prognosis - as well as for investigation into the impact of the disease on the host metabolism. Sebum sampling has the potential to support both needs by looking at what the virus does to us, rather than looking for the virus itself. In this pilot study, sebum samples were collected from 67 hospitalised patients (30 COVID-19 positive and 37 COVID-19 negative) by gauze swab. Lipidomics analysis was carried out using liquid chromatography mass spectrometry, identifying 998 reproducible features. Univariate and multivariate statistical analyses were applied to the resulting feature set. Lipid levels were depressed in COVID-19 positive participants, indicative of dyslipidemia; p-values of 0·022 and 0·015 were obtained for triglycerides and ceramides respectively, with effect sizes of 0·44 and 0·57. Partial Least Squares-Discriminant Analysis showed separation of COVID-19 positive and negative participants with sensitivity of 57% and specificity of 68%, improving to 79% and 83% respectively when controlled for confounding comorbidities. COVID-19 dysregulates many areas of metabolism; in this work we show that the skin lipidome can be added to the list. Given that samples can be provided quickly and painlessly, we conclude that sebum is worthy of future consideration for clinical sampling. The authors acknowledge funding from the EPSRC Impact Acceleration Account for sample collection and processing, as well as EPSRC Fellowship Funding EP/R031118/1, the University of Surrey and BBSRC BB/T002212/1. Mass Spectrometry was funded under EP/P001440/1.

The rapid emergence of antibiotic resistant bacterial pathogens constitutes a critical problem in healthcare and requires the development of novel treatments. Potential strategies include the exploitation of microbial social interactions based on public goods, which are produced at a fitness cost by cooperative microorganisms, but can be exploited by cheaters that do not produce these goods. Cheater invasion has been proposed as a ‘Trojan horse’ approach to infiltrate pathogen populations with strains deploying built-in weaknesses (e.g., sensitiveness to antibiotics). However, previous attempts have been often unsuccessful because population invasion by cheaters was prevented by various mechanisms including the presence of spatial structure (e.g., growth in biofilms), which limits the diffusion and exploitation of public goods. Here we followed an alternative approach and examined whether the manipulation of public good uptake and not its production could result in potential ‘Trojan horses’ suitable for population invasion. We focused on the siderophore pyoverdine produced by the human pathogen Pseudomonas aeruginosa MPAO1 and manipulated its uptake by deleting and/or overexpressing the pyoverdine primary (FpvA) and secondary (FpvB) receptors. We found that receptor synthesis feeds back on pyoverdine production and uptake rates, which led to strains with altered pyoverdine-associated costs and benefits. Moreover, we found that the receptor FpvB was advantageous under iron-limited conditions but revealed hidden costs in the presence of an antibiotic stressor (gentamicin). As a consequence, FpvB mutants became the fittest strain under gentamicin exposure, displacing the wildtype in liquid cultures, and in biofilms and during infections of the wax moth larvae Galleria mellonella, which both represent structured environments. Our findings reveal that an evolutionary trade-off associated with the costs and benefits of a versatile pyoverdine uptake strategy can be harnessed for devising a Trojan-horse candidate for medical interventions.

The last 3 years have seen an explosion in published manuscripts analyzing open-access health datasets, in many cases presenting misleading or biologically implausible findings. There is a growing evidence base to suggest that this is due in part to artificial intelligence-assisted and formulaic workflows, and publishers are responding by discouraging submissions employing open-access health datasets. Here we use a scientometric analysis to investigate which datasets have seen publication rates deviate from previous trends, especially where this coincides with changes to author geographical origins and increases in formulaic titles. Across 36 datasets, we identify nine showing hallmarks of paper mill exploitation (FDA Adverse Event Reporting System, National Health And Nutrition Examination Survey, UK Biobank, FinnGen, the Global Burden of Disease Study, Medical Information Mart for Intensive Care, China Health and Retirement Longitudinal Study, Centers for Disease Control and Prevention Wide-ranging Online Data for Epidemiologic Research, and TriNetX). These nine datasets had, in 2025, a combined publication count of 23,005 indexed in the OpenAlex database. This represents an excess of 11,577 publications above the AutoRegressive Integrated Moving Average forecast trend, and is a 3.0×-fold change on the 7655 publication count for these nine datasets in 2022. We also identified a notable difference in the fold change for China (4.2×) vs. the rest of the world (1.9×) and an increase in formulaic titles. These findings highlight potential risks to research integrity in areas such as public health and drug safety, and especially to the accessibility and interoperability principles central to Open Science and Findable, Accessible, Interoperable and Reusable data practices. We argue that permissive open-access data policies naturally facilitate exploitative workflows and that these findings add to the case for the safeguarding mechanisms to preserve the goals of Open Science.

Identification of features with high levels of confidence in liquid chromatography–mass spectrometry (LC–MS) lipidomics research is an essential part of biomarker discovery, but existing software platforms can give inconsistent results, even from identical spectral data. This poses a clear challenge for reproducibility in biomarker identification. In this work, we illustrate the reproducibility gap for two open-access lipidomics platforms, MS DIAL and Lipostar, finding just 14.0% identification agreement when analyzing identical LC–MS spectra using default settings. Whilst the software platforms performed more consistently using fragmentation data, agreement was still only 36.1% for MS2 spectra. This highlights the critical importance of validation across positive and negative LC–MS modes, as well as the manual curation of spectra and lipidomics software outputs, in order to reduce identification errors caused by closely related lipids and co-elution issues. This curation process can be supplemented by data-driven outlier detection in assessing spectral outputs, which is demonstrated here using a novel machine learning approach based on support vector machine regression combined with leave-one-out cross-validation. These steps are essential to reduce the frequency of false positive identifications and close the reproducibility gap, including between software platforms, which, for downstream users such as bioinformaticians and clinicians, can be an underappreciated source of biomarker identification errors.