Connecting the dots within Wellcome Collection

How can cultural institutions use technology to build a more equitable future? Wellcome Collection are transforming digital access to their vast holdings through a new Knowledge Graph – connecting people, subjects, and works like never before.

Step inside the Wellcome Collection in London and you’ll quickly realise this isn’t a traditional museum. It’s a place where art collides with science, where medieval birth scrolls sit alongside zines, and where conversations about health stretch from 14^th century medical books to contemporary art.

Now, the museum and library is reimagining how its vast digital holdings can be explored. Behind the scenes, a multidisciplinary team is building a knowledge graph—an ambitious attempt to connect people, subjects, and works in ways that prompt wider, deeper discovery and surface hidden content.

“Our vision is a world where everyone’s experience of health matters,” explained Jonathan Cates, Technical Product Manager at Wellcome Collection. “To achieve that, we need to open up our collections to as many people as possible.”

A Collection Without Boundaries

The scale is formidable: over one million works searchable online, 55 million digitised pages from 250,000 books, and a visual archive of 250,000 prints, paintings, and photographs. The holdings are global in scope, spanning more than 50 languages and ranging from centuries-old manuscripts to contemporary ephemera.

“We know our audience is diverse,” said Cates. “Researchers, students, journalists, curators, and artists all use our collections to create new perspectives on health and human experience.”

Yet even with powerful search tools and dedicated web pages for the themes occurring in the collection, many users struggled to make connections between works, subjects and people. “These theme pages weren’t connected or discoverable.” Cates admitted. “In effect, we were pushing people into dead-ends.”

From Dead Ends to Discovery

The team decided to start with an in-depth research phase. “We needed to make sure we were solving real problems for real people,” said Lauren Baily, Product Manager for Digital Experience.

Through surveys, interviews, pop-up testing, and behavioural data, they uncovered a common pattern: researchers sensed there was more in the collection but didn’t know how to reach it. Users often typed in broad search terms, hoping for an overview, and craved a way to browse, stumble upon, and follow unexpected paths.

One clue lay in an existing feature: a tool powered by machine learning that surfaces visually similar images. “It is hugely valued,” Baily noted. “It showed us that users want exploration, spontaneity, and incidental discovery.”

Enter the Knowledge Graph

One of the team’s solution was to build a knowledge graph: a data model that connects works, people, and themes through structured relationships.

“With a knowledge graph, we can capture and enrich information that connects people, organisations, and subjects,” Cates explained. “By integrating external data sources—like medical subject headings, the Library of Congress, and Wikidata—we can enrich our own data and open entirely new routes into the collection.”

Graphs excel at revealing hidden clusters, predicting missing links, and powering smarter recommendations. And crucially, they can be visualised, turning metadata into maps of ideas.

Building Together

Developing this system required a collaborative approach. Two in-house product teams—engineers, designers, analysts, and researchers – with input from internal subject experts.

“All team members are involved in ideation, research, and discovery,” Baily said. “And through workshops, we’ve harnessed the deep knowledge of our expert colleagues.”

The results are now live: enhanced theme pages that transform static listings into rich hubs. A page on Florence Nightingale, for instance, separates works in the collection she authored from those about her, while also pointing users toward related or broader themes. Instead of a catalogue of individual works, the collection has become a network of ideas on health past, present and future.

Tackling the Challenges Ahead

But the work is far from finished. Inconsistent metadata has led to duplication—for example several “Florence Nightingales” instead of one unified entry. “We’ve been experimenting with large language models to identify matches,” Cates explained. “If we can get this right, we’ll eliminate duplicates and enrich theme pages significantly.”

The team is also rethinking the collections landing page. “Right now, users see a search bar but little sense of what’s inside,” Baily said. A new version will showcase popular topics, techniques, and stories, giving users a sense of the breadth, depth and diversity of the collection and a way to start exploring without needing to know the perfect keyword.

Toward an Expansive Future

What emerges is a collection that can be browsed as well as searched; where users can follow a thread or go down a rabbit hole, creating their own paths of exploration. Instead of narrowing research, it opens new routes.

“Our data tells us that theme pages help users discover more works,” Baily concluded. “By surfacing connections, we can spark inspiration and open unexplored avenues of research.”

For museum professionals, the Wellcome Collection’s experiment with knowledge graphs offers an insight into how combining user-centred design, technical innovation, and multidisciplinary collaboration can create new possibilities for digital discovery.

–

Jonathan Cates and Lauren Baily spoke at the Digital Collections Summit 2025.