How to counteract the delegitimization of science

Accuracy, reproducibility, and the appropriate use of controls constitute core pillars of experimental science. Public trust in scientific research is grounded in the assumption that scientists conduct their work with the highest possible accuracy, apply rigorous experimental controls, perform validated data analyses, and provide sufficiently detailed descriptions of materials and methods. Such transparency enables independent experts to reproduce, validate—or, if necessary, refute—the reported findings. In recent years, the public has increasingly been exposed to so-called “alternative facts,” some of which are propagated even within academic contexts. These claims often originate from ideology, which relies on belief rather than rational, evidence-based analysis. Galileo Galilei stands as a historical example of a scientist who suffered from an ideology-based alternative reality, particularly when enforced by powerful institutions. Today, alternative facts and misinformation are disseminated by politicians and interest groups, as exemplified by climate change denial,1, 2 posing a serious challenge to scientific credibility. Generative artificial intelligence may play an ambivalent role in this regard, serving both as a tool for fact-checking3 and as a means for amplifying misinformation4, 5 if misused. Importantly, disagreement and controversy are not pathologies of science but essential drivers of scientific progress. Competing hypotheses are resolved through continued expert discourse and, ultimately, through the generation of additional experimental evidence. As emphasized by Karl Popper, science does not claim possession of truth but rather pursues an ongoing quest for truth.6 The HIV/AIDS saga provides a sobering illustration of this process. In the 1980s, Peter Duesberg, a prominent retrovirologist at the University of California, Berkeley, questioned the causal role of HIV in AIDS.7 Ultimately, isolation, visualization, cloning, and sequencing of HIV enabled the development of effective antiretroviral therapies and established the causal relationship beyond reasonable doubt.8, 9 This stands in stark contrast to modern science denialism—such as that propagated by Robert F. Kennedy Jr. regarding vaccines—which often rejects the fundamental rules of evidence and scientific consensus itself, rather than engaging in the productive, evidence-based discourse that resolved the HIV/AIDS debate. Alternative facts, however, need not arise solely from ideological motivations. They may also result from inadequately conducted science, which is the focus of this article. Such work differs from overt junk science10 in that it often stems from insufficient adherence to fundamental principles of experimental design, data analysis, integration, and validation. Accuracy is not an absolute parameter but a graded concept, and individual researchers may differ in their judgment of what level of precision is sufficient to support a given conclusion. Editors and reviewers therefore face the recurring challenge of evaluating limitations in accuracy across submitted manuscripts. One frequent issue is selective data presentation. The widespread practice of condensing results into multi-panel figures has often led to excessive cropping of gel-based experiments, such as Western blots or electrophoretic mobility shift assays. When only a single signal band is shown without the corresponding background or input controls from the same lane, meaningful interpretation becomes impossible. Recognizing this problem, many journals now require the submission of uncropped gel images, both to confirm the integrity of signal and control lanes and to detect potential image manipulation. A further critical aspect concerns the validation of nucleotide sequences used in molecular biology. Most contemporary studies rely on PCR-based amplification, RNA interference, or related technologies, all of which depend on the accurate design of primers or si/shRNA sequences. Yet, due to time constraints or limited expertise, sequence accuracy is rarely verified during peer review. Alarmingly, editorial analyses have revealed that in multiple cases, sequences were incorrectly designed or insufficiently specific, thereby undermining the validity of the reported results. In one notable case, an improperly designed siRNA was described in a patent and subsequently reused across multiple publications without independent verification. This underscores the responsibility of editors, reviewers, and corresponding authors to critically evaluate sequence design. Statistical rigor is an additional component of accuracy. Appropriate statistical power analyses, transparent reporting of sample sizes, and the inclusion of confidence intervals are essential for assessing the robustness of experimental conclusions. Nevertheless, deficiencies in statistical reporting remain widespread across the biomedical literature.11 The prevailing “publish or perish” culture further exacerbates these problems. The pressure to publish frequently encourages quantity over quality and contributes to a proliferation of underpowered or insufficiently validated studies. Although initiatives such as the Declaration on Research Assessment (DORA) seek to counterbalance these incentives, their impact remains uneven. The rise of predatory journals and paper mills further amplifies the problem.10 Reproducibility represents a central criterion for scientific validity. Only observations and conclusions that can be independently reproduced should be regarded as reliable. A well-known example is the 1988 Nature article12 reporting the so-called “memory of water.” Although the original findings were claimed to be reproducible across several laboratories, subsequent blinded and randomized experiments failed to confirm the effect. The most plausible explanations involved inadequate experimental design and observer bias. Despite this, the article was never formally retracted. Less spectacular but more pervasive is the problem of insufficiently detailed Materials and Methods sections. Authors frequently omit crucial details or refer to prior publications, which themselves lack methodological clarity. This cascading loss of information renders faithful replication nearly impossible. Indeed, surveys indicate that a majority of researchers have failed to reproduce published experiments, including their own.13 Such irreproducibility poses a serious threat to scientific credibility and public trust. Proper controls are indispensable for meaningful experimentation. As emphasized to me by my scientific mentor, Pierre Chambon, experiments lacking adequate controls are scientifically worthless. Control experiments are often complex and demand a high degree of critical self-assessment. Depending on experimental complexity, multiple layers of controls may be required. RNA interference experiments illustrate this challenge well. The common use of scrambled si/shRNAs as controls fails to address well-documented off-target effects. Additional validation strategies, such as rescue experiments or orthogonal technologies including CRISPR/Cas-based genome or transcriptome editing, are therefore essential. Each of these approaches, however, introduces its own limitations, necessitating further controls. Advances in high-throughput technologies have led to an exponential increase in publicly available datasets. For example, by 2024 the GEO database contained data from over 200,000 studies and more than 6.5 million samples, an amount that doubles every ∼5 years.14 While these resources offer unprecedented opportunities for hypothesis generation, they have also fueled an explosion of retrospective biomarker studies with limited clinical relevance. Despite the publication of tens of thousands of prognostic gene expression signatures, only a small number have received regulatory approval for clinical use. The primary limitation is the lack of prospective clinical validation. Consequently, some journals, including the International Journal of Cancer, now apply particularly stringent criteria to studies proposing prognostic signatures derived solely from public datasets. In summary, counteracting the delegitimization of science requires renewed commitment to accuracy, reproducibility, appropriate controls, and meaningful validation.