Heard at HLTH: Can we get Ironman’s J.A.R.V.I.S. in the world of drug development?

Ironman aficionados know that J.A.R.V.I.S  (Just A Rather Very Intelligent System) is the artificial intelligence system that the movie’s protagonist — Tony Starnk —  created to not only control his suit, but his home as well as businesses.

On Sunday, the J.A.R.V.I.S reference made its way to a panel on AI in drug development  at the annual HLTH conference that kicked off in Boston after being run virtually in 2020. One panelist, Miruna Sasu, chief strategy officer at Cota, declared that everyone is hunting for a J.A.R.V.I.S. like system to solve every problem.

“All of us are looking at AIb and thinking, ‘When can we get that? When can we get J.A.R.V.I.S. in every use case?’ ” she asked. “So I am not sure that’s doable in a lot of use cases. It’s not J.A.R.V.I.S.  necessarily. It might be something that helps scientists make sense of a dataset that 20 years ago would have never come together, and they would have never been able to see patterns in the data because we could never spend the time to even put it together.”

As an example of narrower goals AI could achieve, Sasu said that AI could have real impact in finding patients for clinical trials and matching them to do it.

Shahram Ebadollahi, Sasu’s co-panelist and chief data and AI officer with drug giant Novartis, couldn’t agree more.

“People gravitate to the very very hard problem … but I think the biggest prize is embedding AI in small cogs, in small engines in every part of your operation — no matter if it is discovery, no matter if it is development, no matter if it is manufacturing or even commercial activities, that is the biggest value of AI,” Ebadollahi declared  “How it affects us in our daily lives, is through those cogs.”

However, not every panelist agreed with taking a narrower approach to applying AI in the world of drug development.

“I actually believe the early scientific phases is very amenable to AI,” said Daphne Koller, founder and CEO of insitro. But she added that making AI work in this early stage of drug development necessarily requires the tech to work in partnership with scientists.

While perspectives differed on where AI could be most useful, the panelists were unified in their view that the data on which AI algorithms are trained, are woefully inadequate.

“There is a lot of data that is not usable at all so that needs to be first sorted out,” Ebadollahi said.

Koller was more specific on the “significant data quality issues that she faces. For example, the data is based on a certain population that may not be ethnically diverse, and the AI is based off on that data and applied to a much wider and diverse population set including people of color. The result of such extrapolation leads to “wrong results,” she said and could be downright dangerous.

Dealing with different kinds of data that emerged during Covid was a challenge that Carolyn Magill, CEO of Aetion, tackled. The real world evidence company was working with FDA to understand the safety and effectiveness of Covid diagnostics and potential treatments  Covid in the early days of the pandemic. She said that only a nuanced approach to the data that they were getting and reviewing was key in understanding it.

Data is the lynchpin on which application of AI hinges.

“I think we need to invest in data curation specially in the United States, because I think data is really complex.” Sasu said.

Photo: metamorworks, Getty Images

You may also like...

Leave a Reply

Your email address will not be published. Required fields are marked *