Fast company logo
|
advertisement

FUTURE OF HEALTH

How ‘digital twins’ could change the way we develop new drugs

Unlearn.ai’s technology aims to shorten drug approval time—and improve the ethics of clinical trials.

How ‘digital twins’ could change the way we develop new drugs

[Source photos: nonmim/Getty Images;
Towfiqu barbhuiya
/Unsplash]

BY Adam Bluestein4 minute read

Clinical trials are the quicksand of drug development. Clinical trials to support a new drug can drag out for six to seven years, on average, with a median cost of $19 million. Now, using machine learning and generative AI, a startup called Unlearn.ai is aiming to transform clinical trials by tackling one of its biggest challenges: recruiting enough people to participate. Its solution? Digital twins.  

Finding qualified participants for clinical trials—people who match specific disease characteristics and target demographics—has long been a challenge, especially when drugs reach critical Phase 3 studies, which can require hundreds or thousands of participants to test an investigational drug. Clinical trials of new drugs typically compare an investigational treatment to the current standard of care—or a placebo, if there is no standard-of-care therapy that’s approved or available. 

In random controlled trials, people who enroll in the trial are randomly placed into one “arm” or another—the experimental arm or the control arm. The efficacy of a treatment is determined by the difference in outcomes between the groups. Often patients who sign up for experimental treatments are desperate cases, and do not want to be in the control group. Ethically, there’s growing sensitivity to enrolling patients in trials only to give them standard-of-care treatment, which is already available without the additional hassles that participating in a trial may require. 

The concept of a digital twin has been around for a while: It’s basically a software simulation of some real-world system—a jet engine, a factory, a global supply chain—used to predict how it will run, and how it will break down, over time. Although widely used in systems engineering and fields like manufacturing, the digital twin is a relatively new concept in life sciences. “People in the healthcare space think it must be a healthcare concept, because it has the word ‘twin’ in it,” says Unlearn.ai’s CEO and founder Charles Fisher, who previously worked as a machine learning engineer at Leap Motion and as a computational biologist at Pfizer. 

The digital twins that concern Fisher at Unlearn—which now employs about 60 people and this year raised $50 million in funding—are individualized computer simulations of people. “We’re taking historical data, real-world data, about a particular disease and how it progresses on current treatments, and encapsulating that within a computer model,” he says. “Then, if I enroll in a clinical trial, we take data from me at the beginning of the trial, put that into the model, and it creates simulations of what might happen to me in the future.” So, if the real you gets the experimental treatment, your digital twin gets run through an algorithm to see what would have happened if you didn’t. 

The technology underlying Unlearn’s platform is a combination of classical machine learning and so-called generative AI, of which DALL-E 2 is probably the most famous current example. “We are taking the same kinds of models that they are using for generating images and applying it to our work generating synthetic patients,” Fisher says. How accurate is it? Unlearn has had an early focus on Alzheimer’s disease, and in a peer-reviewed September 2019 article in Nature Scientific Reports, Fisher and his team reported on the development of digital twins for Alzheimer’s disease that generated detailed synthetic patient data on disease progression over 18 months that was statistically indistinguishable from the actual data.  

New digital twin models will be validated through a hybrid trial design, Fisher says. Study participants will still be randomized into a treatment and a control group. But thanks to the added power of digital twin predictions, the human control groups can be up to a third smaller. At the beginning of the trial, data from all participants will be used to create twins that simulate the outcome of treatment with a placebo. Actual patients randomized to the placebo group will provide an internal measure of how accurate the model is in the actual trial population. “We’re making the trials smaller only through making the control group smaller,” says Fisher. “So, if you participate in one of these trials, you have a much higher probability of getting the experimental drug.”

Unlearn is working with pharmaceutical industry partners to test potential treatments for neurodegenerative diseases, says Fisher. In February 2022, the company announced a multiyear partnership with Merck KGaA, the German pharma giant that is a separate entity from U.S.-based Merck & Co., to focus on immunology products. This September, the European Medicines Agency said it would allow data derived from Unlearn’s digital twins platform to reduce clinical trial sizes in Phase 2 and Phase 3 clinical trials with continuous outcomes (i.e., long-running trials that are measured at multiple follow-up points).

In the U.S., says Fisher, the Food & Drug Administration does not yet have any process to get a qualification or approval or review of an approach like Unlearn’s. “Effectively, we’re kind of unregulated, I guess,” he says. “When you submit your clinical trial protocol before you run your trial, there’s a page that describes what we’re doing, and the FDA signs off on your protocol on a case-by-case basis.” 

That may not be enough to build a business on, though. “You’re not going to blow $10 million on digital twins if you can’t use the data for your trial,” says Derk Arts, founder and CEO of New York- and Amsterdam-based Castor EDC, a platform for managing decentralized clinical trials. But digital twins can still inform in-house R&D and clinical trial design, he says. “I’d be surprised if every major pharma company doesn’t have one or more people looking at [digital twins], to try to have an early idea of how a trial is going to run.” 

Recognize your company's culture of innovation by applying to this year's Best Workplaces for Innovators Awards before the final deadline, April 5.

PluggedIn Newsletter logo
Sign up for our weekly tech digest.
This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.
Privacy Policy

ABOUT THE AUTHOR

Adam Bluestein writes for Fast Company about people and companies at the forefront of innovation in business and technology, life sciences and medicine, food, and culture. His work has also appeared in Fortune, Bloomberg Businessweek, Men's Journal, and Proto More


Explore Topics