Digital Reasoning, a data-analytics software firm based in Brentwood, Tenn., is looking to move beyond its core market in the defense and intelligence community into the life-science and healthcare sector.
The company recently signed a collaboration with Battelle Science and Technology International to expand its core technology, which automatically extracts relationships from unstructured data, into additional intelligence applications as well as new applications in the life sciences.
“Our focus has been in the defense and the intelligence area, but we’ve had an active interest in the healthcare space along several different vectors,” Tim Estes, Digital Reasoning’s CEO, told BioInform this week.
The company claims that its software searches natural language texts and automatically identifies the contextual usage of terms to extrapolate semantic relationships.
“We came in to solve the problem of not so much connect the dots, because the people who are in the [defense and intelligence] space will tell you that if they know what the dots are, the technologies are pretty much there to connect them if you can get the data,” Estes said. “The problem is you don’t know what a dot is until after the fact.”
While this capability has obvious applications in the intelligence community, Estes said that Digital Reasoning has identified several use cases within the healthcare sector where it might also be of use.
Initially, the company sees a role for its technology in patient records, which could be mined to more quickly identify severe adverse events or secondary indications for drugs on the market.
“If we can succeed in coming up with an automated way to detect these severe adverse effects directly from the doctor/patient record, which is our goal, and to come up with secondary indications that would provide benefits to the providers of drugs and devices, then we would like to see whether we can actually learn enough to optimize protocols,” Estes said. “Can we learn enough to make treatments best fit not just assumptions about the market, but actually best fit assumptions about phenotypic expression in a given person?”
Estes noted that the long-term goal would be to support personalized medicine. “The idea of broad efficacy that’s pretty similar across all people is actually a fiction,” he said. “Most of the interesting work is how to figure out what type of person a drug really works well on, and what classes it doesn’t work well on.”
Estes doesn’t see a role for Digital Reasoning’s technology in analyzing molecular-scale information, which is typically structured. Nevertheless, he said, the company envisions “a composite solution being on the horizon.”
“Can we learn enough to make treatments best fit not just assumptions about the market, but actually best fit assumptions about phenotypic expression in a given person?”
By integrating structured molecular data with phenotypic information captured in medical records data, physicians would have more “variables” to determine why certain symptoms might occur for one person, and not another. “Given that their phenotypic expression is fairly similar, maybe it’s a genetic issue,” he said.
Estes stressed that the company’s plans for the life science market are long-term, and will largely depend on initiatives underway in the US to develop a nation-wide network of electronic health records.
“We don’t have a direct product offering or foray into that area yet because the market is just catching up infrastructure-wise to make it possible,” he said. “So we’re looking to take the people who are very progressive in those areas and work with them as partners.”
Estes said that translating the company’s technology into the healthcare market should be relatively straightforward because medical information is not as noisy as data in the defense and intelligence sector.
“The training of physicians is so effective, from the standpoint of there being well-structured knowledge in the educational process, that as they capture certain symptoms and impressions, it’s already semi-structured because of the nature of the knowledgebase,” he said.
“We’re dealing with things where the concepts aren’t clear, there’s no common vocabulary, there are words entering the language all the time that didn’t exist before,” Estes said. “So that’s kind of a harder problem than a marketplace where the terminology has to stay somewhat static or related to each other in a strong way so that people can communicate well.”