The National Institutes of Health's National Library of Medicine has released the pilot version of a new tool designed to help scientists plumb the library’s digital trove of scientific papers.
Called Linked Discoveries, the new feature lives on PubMed, NLM’s database of biomedical references, and is part of NIH Director Jay Bhattacharya, M.D., Ph.D.’s broader mission to improve reproducibility in research.
“When scientists can more readily test, verify, refine and build upon previous work, we create firmer ground for the discoveries that follow,” Bhattacharya said in a call with journalists in advance of the feature’s release. “That is science working as it should, continually testing itself, correcting itself, becoming stronger as a result, and Linked Discoveries is one part of NIH's larger effort to make that possible.”
In a demo of the feature shown to journalists, a button to access Linked Discoveries sat on the right side of the PubMed page for a chosen scientific paper. Clicking the button pulled up a graphic showing how that chosen paper connected to a so-called “neighborhood” of other papers. This includes papers that cite the focal paper and vice versa, as well as visual indicators that help show researchers the relative influence of each paper and whether certain topics are mentioned in them.
“Linked Discoveries was born from a call to create a replication button on PubMed, and this was a bold idea to make it easier for people to determine whether a paper's findings have been tested,” Stephen Sherry, Ph.D., NLM’s deputy director for scientific operations and innovation, said during the same briefing.
The tool is designed to help combat the fact that “highly cited findings may receive disproportionate attention, while contradictory, null or otherwise less prominent findings can be easier to miss,” Sherry added.
Linked Discoveries gives researchers using PubMed “a broader view of the work surrounding a study, what came before it, what came after it, and what other evidence might be important to consider,” NLM Director Peter Embí, M.D., said during the briefing.
The new PubMed feature is built on a publicly available language-processing model called BiomedBERT, which differs from the large-language models that are currently dominating the AI conversation, according to Sherry.
“This is not a closed development system,” Sherry said. “We're trying to do this in a real-time engine that is constantly recalculating the relationships daily as new articles come in.”
With the tool still in a pilot phase, the NLM is welcoming all feedback from users so that improvements can be made over time.
Other efforts to improve reproducibility in biomedical research, a long-standing concern in the field, are also underway at the NIH, Bhattacharya said. This includes a journal to specifically publish papers attempting to replicate past results.
“This Linked Discoveries button ... is actually very radical,” the NIH director said. “Scientific crosstalk,” between researchers and ideas “is a necessary step for big advances to be made and verified,” he added.