By Betty Zou
Daniel Drucker (MD ’80, PGME ’84) is at the pinnacle of his career. Glucagon-like peptide-1 (GLP-1) medicines, such as Ozempic and Wegovy, are everywhere and are having a massive cultural moment thanks in part to Drucker, a professor at the University of Toronto’s Temerty Faculty of Medicine. Starting in the 1980s, he did the yeoman’s work that underpinned the development of these drugs.
Drucker, who started his scientific career more than 40 years ago, thinks we are currently living in a golden age of science. He also believes science is in crisis.
On both counts, he’s not wrong.
Thanks to decades of research advances, cellphones that double as powerful hand-held computers are now ubiquitous, and once-fatal diseases have transformed into manageable chronic illnesses. And, for better or worse, artificial intelligence (AI) is changing the way we work and live our lives.
And yet, science — and the people doing it — face growing threats from both outside and within the scientific community. Ideas that have been widely accepted as scientific consensus — that climate change is caused by human activity, that vaccines do not cause autism, that HIV causes AIDS — are being challenged because as a society, we can no longer agree on what constitutes a truth.
Scientists, health care professionals, learners and the institutions that support them are called upon to defend their work while also acknowledging and addressing the shortcomings of the current system.
What’s at stake in this fight? Credibility. Progress. Everything.
The righteousness of reproducibility
“My whole life, I have never wanted to be wrong,” says Drucker, a professor in Temerty Medicine’s Department of Medicine and a senior scientist at Sinai Health’s Lunenfeld-Tanenbaum Research Institute. “I know that sounds like the dumbest thing because no one wants to be wrong. But in science, you must bend over backwards to get it right because it’s an obligation we have to the people who are funding the work.”
What makes Drucker unique is not his commitment to careful, curiosity-driven science, but the lengths to which he is willing to go to ensure the reproducibility of findings from both his lab and the larger field of diabetes research, where his work on GLP-1 has had the biggest impact.
When someone in his lab shares a particularly exciting result with him, he often asks another lab member to try and independently replicate the experiment. He’s also known to call out colleagues and journal editors on X (formerly Twitter) when he spots something that’s not quite right — a privilege that comes with his seniority and reputation in the field, and one that has cost him friends and made him enemies.
As a clinician scientist, Drucker is motivated by his patients to be laser-focused on translating discoveries made in the lab into new treatments that improve human health. His pioneering work on the gut hormone glucagon laid the foundation for two new classes of drugs for type 2 diabetes and obesity, and a new treatment for short bowel syndrome.
He sees irreproducibility — across biological traits such as sex and age, from animals to humans, or between labs — as a major challenge that’s slowing down progress not just in his field, but in biomedical research at large. There have not been a lot of comprehensive studies examining the degree of reproducibility across biomedical fields, but the data that do exist are disheartening.
In 2021, a massive eight-year collaborative effort to replicate experiments from high-impact cancer biology papers found that just 46 per cent of replications successfully reproduced the results of the original experiment, a finding in agreement with earlier results from a similar reproducibility project in psychology. When the results did replicate, the effect size was significantly smaller than what was originally reported — half the magnitude in psychology and 85 per cent smaller in cancer biology.
More encouraging results from a retrospective analysis of 400 studies related to immunity in the Drosophila fruit fly found that 61 per cent of claims were verifiable. The findings were shared in two preprints posted to bioRxiv in July 2025.
Reproducible Open AI starts with open code
As a relatively young field, there is scant data on reproducibility in AI research, particularly as it applies to health and medicine. But that has not deterred Benjamin Haibe-Kains from calling for more transparency and openness in sharing the codes and data sets that are key to developing AI tools.
“One of the core principles of science is that it can be repeated, independently scrutinized and more importantly, reused and improved upon,” says Haibe-Kains, a senior scientist at University Health Network (UHN) and professor of medical biophysics at Temerty Medicine. “If you’re the only one who can generate a result, it has little to no value to humankind,” says Haibe-Kains, who is also the executive AI scientific director and co-director of the AI Hub at UHN.
His research aims to develop computational tools and predictive models that can leverage large data sets to improve our understanding of cancer biology and precision medicine approaches for people with cancer.
In 2020, Haibe-Kains and his colleagues wrote a commentary about a study led by a team of researchers at Google Health and DeepMind describing an AI system they created to interpret mammograms for breast cancer screening. The findings, which were published in Nature, claimed that the AI system was faster and more precise than the work of radiologists, outperforming human experts in certain settings.
While Haibe-Kains and his co-authors did not dispute the system’s potential, they wrote that “the absence of sufficiently documented methods and computer code underlying the study effectively undermines its scientific value.” In response, the study authors expanded the article’s supplementary methods with more, though still not complete, details.
If you’re the only one who can generate a result, it has little to no value to humankind
Truth detectives
Many of the factors that contribute to an experimental finding being irreproducible are the result of what science integrity sleuth Dorothy Bishop calls “questionable research practices.” These include inappropriate statistical analysis (known as p-hacking), selective reporting, insufficient controls and small sample sizes.
“That’s not fraud,” says Bishop, who is an emeritus professor of developmental neuropsychology at the University of Oxford in England. “It’s people not doing things as they should because often, it’s become normative, and they haven’t been trained on how to do things properly.”
Like most of the sleuths she knows, Bishop fell into the hobby by accident in 2015 when she was tipped off about dubious activities at a journal. Initially, she investigated cases of questionable research practices, but over the past five years, she has seen a precipitous increase in what she describes as “industrial-scale fraud” in the scientific publishing industry.
These include paper mills that churn out low-quality, AI-generated papers where for a price, anyone can be listed as an author, and predatory journals that charge authors thousands of dollars to publish a paper with the promise of fast and less-than-rigorous peer review. More recently, the work of Bishop and her fellow sleuths has uncovered what she calls “review mills,” coordinated networks of reviewers who provide generic and often fake reviews to coerce authors to cite the reviewers’ own papers.
The not-so-glamorous side of science
While the line between sloppy science and deliberate misconduct can sometimes get blurred, these actions are all driven by the same unrelenting pressure to publish. Job appointments, promotions, awards, funding, university rankings — so much of it hinges on how often you publish, where you publish and how many other people cite your publications.
Making matters worse, federal spending on research in Canada has fallen short of both inflation and the demand for innovation. According to data from the World Bank, Canada is the only country in the G7 to have had its gross domestic spending on research and development decline between 2002 and 2022. During that same timeframe, funding success rates at the Canadian Institutes of Health Research were cut nearly in half from an average of 31 per cent in 2000–2003 to 17.5 per cent in the 2022–2025 competitions.
The pressures of this system are most strongly felt by early-career researchers — postdoctoral fellows trying to land their first academic job and junior faculty working to secure tenure, says Drucker.
It’s a struggle that Jinglin Lucy Xie (PhD ’17) knows all too well. Xie is a postdoctoral fellow at Stanford University in California who has been on the job market for the last several years, seeking a research faculty position. While she has gotten a few interviews, none have led to a job offer. When she looks at the profiles of the candidates who are making it through the hiring process, they often have a paper in one of the three so-called “glam journals” Nature, Cell or Science.
Xie believes that not having a first-author publication in one of these journals puts her at a significant disadvantage. When we speak, she is rushing to put the finishing touches on a resubmission to Nature just days before the due date of her second child.
“I care more about telling a coherent and compelling story of our discoveries than about whether it’s published in Nature, Cell or Science,” she says. Xie’s resolve to stay true to herself and to the science has meant at times pushing back on including data that would have made the story more exciting, but had not been replicated enough times for her to be confident in the results.
Even for Drucker, who received the 2025 Breakthrough Prize in Life Sciences for his work on GLP-1, publication in these top-tier journals was sometimes out of reach. He notes that the three discoveries from his lab that led to new classes of drugs were all published in PNAS, a respectable but less coveted journal.
“We couldn’t get into Nature, Science or Cell because we didn’t have spectacular and cool enough data for those journals,” he says.
I can more about telling a coherent, compelling story
Making enemies and progress
In a system that rewards quantity over quality and favours eye-catching, flashy headlines over meticulous, incremental advances, it’s perhaps not surprising that irreproducibility and fraud have emerged as major threats to scientific integrity.
“You get what you incentivize, and right now, we are incentivizing the wrong things,” says Drucker.
Felix Cheung agrees.
“The current incentive structure in science encourages us to do what’s good for scientists, but not necessarily what’s good for science,” he says.
Metascience — or as Cheung explains it, “the scientific study of how science works” — is trying to change that by bringing the two closer in alignment.
Now an assistant professor of psychology at U of T’s Faculty of Arts & Science, Cheung was drawn to the metascience movement as a young idealistic graduate student. He started graduate school in 2010, when concerns about a reproducibility crisis in psychology were at a peak. That motivated him to sign up to help replicate one of 100 studies tested in the Reproducibility Project: Psychology.
Cheung recalls the lengths to which he and his colleagues went to replicate the original experiment, including accounting for details, such as room layout and décor, that were not in the published manuscript. Despite their best efforts, they were unable to reproduce the results from the original experiment. Their results triggered heated exchanges that played out on X and in science and mainstream media.
Cheung remembers a comment from a professor at Harvard University who said that only second-rate researchers do replication studies because first-rate researchers conduct novel research.
“There were senior researchers at the time who tried to convince me not to pursue metascience research to ensure my own survival in the discipline,” he says. “When I started to make enemies, that’s when I realized that I’m standing up for something important.”
Today, he has found allies who share his commitment to improving science and is encouraged by the progress he’s seen in his field and others. For example, when he joined U of T’s faculty in 2020, he was heartened to see that the reproducibility project was already being taught as part of the undergraduate curriculum in psychology.
More journals and funders are requiring researchers to pre-register their studies, a practice that Cheung and his trainees already follow to reduce bias in reporting and analysis. Some journals are incentivizing reproducibility by publishing replication studies.
The current incentive structure in science encourages us to do what’s good for scientists, but not necessarily what’s good for science
Shifting the culture of science
Every scientist you talk to will have a different idea of what is needed to make science more rigorous and trustworthy.
Drucker advocates for reproducibility to be included as a metric in promotion and funding applications, as well as for the publication of negative results. Haibe-Kains has created tools to help researchers share large biomedical data sets and to help institutions track their open science contributions. But they all agree that fundamentally, the culture within science must change.
“To change that culture, it has to come from putting the right incentives in place,” says Haibe-Kains.
Some of that change has already started. Bishop, who played a key role in establishing the UK Reproducibility Network, says that in the United Kingdom and Europe, funders typically ask for an applicant’s best or most impactful publications, making it easier for researchers to focus on the quality of their work rather than the quantity.
Xie notes that in the United States, all graduate students and postdoctoral fellows at institutions that receive funding from the National Institutes of Health are required to take courses on the responsible conduct of research.
In Canada, the Canadian Institutes of Health Research has mandated a course on research ethics for anyone whose research involves human participants, but there are no other requirements for trainees. Some institutions have stepped in to fill that gap, but the efforts have been piecemeal. Temerty Medicine’s Office of the Vice Dean, Research and Health Education, for example, offers a graduate research integrity workshop.
“Sometimes people step on ethical boundaries because they are naive about certain aspects of research ethics,” says Xie. “These courses increase awareness so that even if their supervisor doesn’t mention it, trainees still think about it.”
Concerns about reproducibility and misconduct in science are not new. What makes this moment different is the politicization of these issues and how they have been weaponized as an excuse to defund universities and strip away support for research across a range of disciplines.
“I feel conflicted, but we have to keep doing it,” says Bishop about her efforts calling out cases of misconduct. She worries that if fraudulent activities continue unchecked, science will become more insular as trust in science deteriorates among both the general public and within the scientific community itself.
Bishop says she is already hearing from researchers who are less willing to trust studies from Russia, China, Iran and other countries where paper mills have reportedly been found. And if the crisis continues, there’s the risk of walking back the progress that’s been made to make science more inclusive and losing a whole generation of talented young researchers, she says.
As for Xie, she is taking a temporary break from the job search to care for her son, who was born at the end of August.
“I’m still hopeful,” she says. “I believe in my science and the story I have to tell. It’s a matter of me convincing other people, and I’m not willing to give up yet.” •