We all want to ensure that scholarly research is efficient, reproducible, reusable, and trustworthy. Shifting the format of scholarly communication from a flat PDF article to a networked package of research outputs including data, code, methods, and other assets, may be the best way to do that. And with editorial support from AI publishing tools like DataSeer SnapShot, that vision of holistic research communication is on the verge of becoming a practicable reality.
Earlier this month, at the NISO Plus conference in Baltimore, presenters Kristen Ratan of Stratos (who is also a member of the DataSeer advisory board) and Jennifer Heimberg of the National Academy of Sciences unpacked the notion as part of a panel discussion entitled Research Stacks and Research Integrity. They explored how alternative forms of research communication can cultivate accountability and research integrity, and help science function more effectively.
The concept of a more complete communication of science, shared by researchers when they feel ready, has been offered as an alternative to traditional publishing for years. The notion of a research stack could augment the process currently in place for preprints.
What does format have to do with integrity?
At first glance, it may not be obvious how the mode of research communication impacts research integrity and by extension, trust, quality, and impact. The matter comes down to using the right tool for the job.
A standard research article is an abridged account of the research process and outcomes. Its purpose is to provide high-level context and summarize findings. It’s a digest — incomplete by design. For that reason, research articles aren’t very good at supporting reanalysis, replication or reuse — practices which form the backbone of research integrity.
That shouldn’t come as a surprise; replication, reanalysis, reuse and verification aren’t what research articles were built for. The article format is a holdover from an earlier era of print communication, a time when space was limited and knowledge networks did not exist, with different research practices, and different communications goals and standards.
As the information needs of the research community have evolved, we’ve tried to evolve the article format alongside them. To improve understanding, increase the speed of innovation, and better enable replication and reuse (and help prevent fraud), we’ve added adjuncts like supplementary material and data files, implemented research integrity policies and badges signifying quality, and attempted to share progress sooner through preregistration and preprints — but with mixed results.
What’s needed, Ratan and Heimberg argue, is an entirely new research communication mechanism. A research output that is engrained in the research production pipeline and leverages the tools and technologies researchers already use in their daily work.
Why research stacks?
There are three main issues preventing articles from being effective in the modern context. First, articles demand a delay in communicating results. Incremental results are stockpiled while researchers accumulate enough information to make up a full article. That is inefficient and inhibits collaboration. Second, once a sufficient body of evidence has been gathered, it must be extracted from the formats in which it was created and slotted into the article template, most often shared as a flat PDF file, which may or may not link to related research components such as data or code. That is clumsy and wasteful. Reformatting strips the work of its complex and eminently usable in-built knowledge networks and isolates it, making it more difficult to work with. Finally, articles don’t reflect how researchers work in the world of cloud-based big data and real-time collaboration.
This figure came from a recent workshop held on the Janelia campus in December 2024, where 40 experts innovating in new forms of research communication came together to imagine what a new stack or container would look like and how it might accelerate research outcomes. This topic will be covered at an upcoming ICOR web meeting on February 27 at 8am PT, 11am ET, 4pm GMT. Register and join the conversation.
Today, most aspects of research, from methods, to data collection and recording, to analysis, take place in a digital environment, producing native documentation that could easily be packaged and shared. What if, as Ratan and Heimberg posit, we combined that into a holistic research package or “stack” containing all the detail necessary to engage with the work on every level — to understand, validate, and build upon it? Researchers would save time and grant dollars spent needlessly transposing their work into article format. Each element of the stack could be attributed to the researchers responsible for its creation. Collaboration would become more seamless and progress more rapid.
The role of AI in supporting the next generation of research communication
Like an article, the research stack would undergo a screening and certification process that incorporates the editorial checks and peer review feedback already in use at journals. But — due to the added complexity, detail, and sheer volume of information in the stack — existing screening methods would need to be augmented with new technology solutions.
Artificial Intelligence and Natural Language Processing tools can work alongside and in support of human experts. These tools are capable of providing flexible yet consistent automated screening and author-facing feedback — identifying signifiers of research excellence as well as potential errors and malfeasance, providing detailed customized feedback, evaluating revisions to ensure that requested changes are in place, and more.
If that sounds too good to be true, consider this: the first generation of sophisticated AI editorial support tools is already in use. DataSeer SnapShot is one early example, an AI integration that scans manuscripts for research integrity and policy compliance. SnapShot works within existing service provider- and journal- workflows, allowing journal staff to rapidly triage manuscripts, moving those that display signals of research integrity into peer review faster, and automatically providing guidance when it’s needed. Pilot programs at KGL, Taylor & Francis, AAAS, Highwire, and others began in 2024, and are currently ramping up. And that’s only the beginning.
What happens when research stacks become standard?
Looking further into the future, Ratan and Heimberg said, it’s possible to imagine a more dynamic and integrated way of working, in which the research stack is publicly available from the beginning of the research process, with assessment, review, validation, readership, and reuse all taking place on a rolling basis within a single, purpose-built environment where conducting and consuming research are seamlessly blended.
But that vision is still a long way ahead. For now, we can take the first steps by leveraging AI editorial support tools to ramp up efficiency, increase consistency, and reduce workload so that journal teams can focus on the work that really requires their expertise.
