Regeneron Uses Voice-to-Text for Science-First Data Digitization Efforts for Animal Research

By Allison Proffitt 

July 9, 2024 | When Regeneron began exploring how to digitize scientific information in lab environments, IT engineers didn’t begin by brainstorming their own ideas. They began by shadowing scientists.  

“We decided to start shadowing our scientists in the lab to understand what they were doing on a day-to-day basis, the issues they were encountering. We had interview sessions and a lot of meetings with them,” said Patrick Leblanc, director of business relationship management and solution partner at Regeneron Pharmaceuticals.  

That science-first approach won the team a 2024 Bio-IT World Innovative Practices Award, sharing the “Informatics to Achieve Operational Excellence” category with two other Regeneron projects.  

This effort was part of Regeneron’s TIDES program or Transform Information with Digital Experimental Solutions. The mission of TIDES is to revolutionize traditional data capture with innovative technology by turning raw data into scientific insights.  

TIDES addresses challenges in all 22 business areas of Regeneron, but in this case, the issues that stood out to Leblanc and his team were the repetitive manual tasks, process variability, and technical debt incurred by in vivo scientists—those working with animal models. “They were writing down things on a piece of paper and having to transcribe it multiple times before it reached the data source,” Leblanc said. “That led to a lot of inefficiencies in labs, spending more time than they should performing those tasks, data quality issues because of the multiple layers of transcription they needed to do, and lack of clear visibility into the data they were capturing in the labs.” 

The TIDES team focused on solving problems of experimental data capture for vivariums—Regeneron has three comprising some 30,000 cages—and other animal labs, particularly with data capture bundles (automating data input from scales or calipers), wearables like headsets that could assist with handsfree data capture for scientists who were working with animals, and voice-to-text platforms. The team is also exploring using Microsoft HoloLens, a virtual reality headset, for both training and remote technical assistance with instrumentation in the labs.  

Voice-to-text was promising from the start. “We started looking a few years back at the voice-to-text platform to allow us to be able to dictate that information accurately and efficiently,” Leblanc explained. “We even tried to develop our own; we tested many different cost applications and we ended up using LabVoice to capture data measurements and metadata.”  

“You can think of this as Siri or even Amazon Alexa for your lab space,” added Kristian Kolakowski, Principal Business Analyst at Regeneron. The problems were simple: scientists were creating temporary records on post-its, their phones, white boards—once even a paper towel—as they worked with animals in vivariums, biosafety cabinets, and isolators. Later they or a research associate would transfer those written or audio records into the digital record.  

“With LabVoice, we can avoid those temporary records, which could lead to data loss—that data not being FAIR and accessible in the long run,” Kolakowski said.  

Digitized data not only saves time and reduces redundancy, but it drives discovery he added. LabVoice offers a design tool to help configure workflows so that data input follows a prescribed scientific process. “We can ensure greater accuracy in those responses because we can almost predict what type of responses those are going to be,” Kolakowski said.  

The benefits include efficiency and productivity gains, data quality gains, focus gains for scientists, and automated ingestion of digitized data into databases. “If you’re working with animals, you can really worry about that animal’s welfare and not on the multitasking that you’re performing within the lab space,” he said.  

Scientist Buy-In 

While it sounds good in theory, the tool would not succeed if scientist didn’t adopt it. And there were some skeptics. “In the beginning I hated it,” reads one of the testimonials included in the entry, though the scientist added, “now I love using it.” The TIDES team made a concerted effort to address users’ concerns and bring them on board. “That’s something we see often,” Kolakowski said. “Our users may be scared to change because they’ve been doing things a certain way for maybe five to ten years.”  

“We work with diverse processes and project personas. We need to tailor the training to every individual that we’re working with to make sure they are equipped to use the technology,” Kolakowski said. “And we’re there with them, every step of the way to make sure that it’s a smooth transition.” 

Kolakowski recommends clearly outlining plans, demonstrating products so everyone can gain familiarity, and documenting all the gains for users to see—both quantitative and qualitative. For instance, using LabVoice takes a bit longer to record empty animal cages individually, but Regeneron can use the complete data in new ways. When the entry was submitted, LabVoice had effectively captured more than 174,000 data points, both automated and manual, which is being used to understand the number of new litters, test subject attributes, and natural breeding trends across facilities. 

One particular area of concern was how a voice-to-text system would handle varying accents on top of scientific language. Kolakowski reported having users who were native Spanish, Arabic, and Mandarin Chinese speakers in addition to English. They carefully tracked the number of corrections users had to do to LabVoice data and shared as the correction rate fell month by month. In the first quarter of 2024, Kolakowski reported a 1.3% correction rate. “That’s about one out of every hundred responses that’s being corrected over time.”  

“Ultimately, our solution has delivered hands-free data capture for our research scientists while providing key insights from the data,” the authors wrote.