Researchably’s New AI Solution Uses NLP To Sift Through Research
By Benjamin Ross
October 4, 2018 | Researchably, a startup in the UC Berkeley SkyDeck accelerator program, announced their natural language processing (NLP) AI solution, which has the capability to read and review scientific articles, and determine to which departments they are most relevant faster, making far fewer errors than expert readers.
The solution, which will be available through subscription service, is currently conducting a pilot with Sanofi.
According to the company’s official announcement, the new commercial solution cuts the time to scan and review a paper from an average of 13 minutes per paper when done by a person, to less than one second.
This solution addresses a critical need for the research community, Researchably co-founder Maciej Szpakowski, told Bio-IT World. Along with fellow co-founders Mauro Cozzi and Przemysław Zientała, Szpakowski started as a student at the University of Southampton where he observed the challenge of analyzing and reviewing scientific journal articles.
“There are about 30,000 research papers that are published in scientific journals every week,” Szpakowski said. “It’s really hard for students and researchers to go through the medical literature. They’re long, complex documents. And so we saw the opportunity to build the technology that would automate some of those processes.”
Researchably’s platform categorizes scientific papers, reads them for specific details, and then sends researchers a custom selection of appropriate papers based on specialty and interests.
The key to Researchably’s AI platform, co-founder Mauro Cozzi said, is its NLP capability, which he says makes the platform not just “tech for the sake of tech”.
“Every line of code we write is informed by the problems we’re tackling,” Cozzi explained. “So we’re constantly engaging with medical experts to try to understand the nuances of how these research papers are written, why it might be hard to screen them.”
Researchably is able to effectively develop its platform’s NLP by tapping into every openly available dataset. It also helps to have exclusive access to pharmaceutical data, which the Researchably team was able to gain in their partnership with Sanofi.
“Through our partnership with Sanofi, we were able to train these algorithms in ways that perhaps others couldn’t,” said Cozzi.
Sanofi’s trial with the Researchably platform is through its medical communications team in China, which is tasked with keeping doctors informed about the latest developments in their fields.
“[Sanofi]’s focus [for the pilot] is on Type 2 diabetes,” said Cozzi. “Everyday they would review between 100-200 scientific articles in order to understand whether the articles were relevant to colleagues for analysis, or whether it was relevant to a key opinion leader (KOL).”
Szpakowski said Sanofi is able to cut the time from 20 minutes spent on each paper to only seconds with Researchably’s solution. The AI solution also brought new or perhaps overlooked content to the team’s attention.
“What we saw was that even with this team of ten people, they were looking only at 20% of the potentially relevant research papers that were coming out,” Szpakowski said. “Our technology helps them look at the full scope of the papers they would miss.”
While Szpakowski and Cozzi are proud of what they’ve brought to the table, they are still looking for ways to improve their AI’s capabilities.
For instance, Cozzi said, Researchably is looking into expanding the solution’s summarization capabilities in the next few years. The solution currently offers quantity; Cozzi wants it to offer quality as well, breaking down the complexity of a scientific paper into more accessible terms.