Speeding up and bringing down the cost of clinical trials would be a huge boon to pharma companies and patients alike.
San Francisco-based startup Unlearn.AI says it has developed proprietary technology that may help accomplish just that. On Monday the company announced it raised $12 million from investors who believe it stands out among the crowded field of tech outfits looking to tap into the huge amount of money spent conducting such studies.
Companies across industries are taking advantage of improvements in machine learning tools and the increasing availability of computational power to analyze huge amounts of data. The process that Unlearn has developed, dubbed DiGenesis, crunches through historical clinical trials datasets and uses that information to create what the company calls refers to as “digital twins”—virtual versions of thousands of real patients.
|Want more AI, machine learning, and drug discovery content? Network and engage with industry leaders online this June. Learn more.|
The term is borrowed from another industry: manufacturing. When building a jet engine, for example, engineers create a virtual representation of the product, too, to optimize its design and enable better performance.
People, of course, are much more complex than planes. But machine learning tools are especially useful in parsing complicated systems, and Unlearn has built disease-specific models that generate control patient data by using demographic information and observations of the same variables over time, such as common lab tests and biomarkers, to generate virtual medical records.
This approach to is intended to shrink the number of patients needed to kick off a trial. It also could make it easier to determine whether the drug works. That’s because measuring a drug’s performance is more straightforward the closer the patient serving as a control—the one in the trial given a placebo—is to the patient who receives the treatment.
“The problem is … I can’t simultaneously give a person a treatment and not give them a treatment,” said Unlearn founder and CEO Charles Fisher in a phone interview. But with digital twins, Unlearn’s pitch proposes, scientists can, to a degree, circumvent that limitation.
Fisher, who launched the company in 2017, was formerly a computational biologist at Pfizer (NYSE: PFE), where he worked to develop machine learning approaches to improve clinical trials. He likened the distinct approach Unlearn has developed to research done by Nvidia (NASDAQ: NVDA), which used a type of neural network—the name for a class of machine learning algorithms modeled after the human brain—to generate endless portraits of fake faces from a huge dataset of real images.
“Most machine learning has been done on images and some on language, but we’re working with medical records,” he said. “We have a generative model that’s trained on previous clinical records, and so then we can generate new clinical records that aren’t exactly the same as any particular person that’s in our database, but which have the same general properties of patients from the control arms of clinical trials.”
Unlearn’s initial focus is on Alzheimer’s disease and multiple sclerosis, neurological indications for which studying potential new treatments is especially expensive and time consuming.
In addition to pharma companies, Unlearn will also have to convince regulators that its approach is effective.
Fisher says the company has had a “lot of good conversations with regulators,” including a Critical Path Innovation Meeting with the FDA in March. The agency describes such meetings as a “forum for FDA and stakeholders to discuss potential scientific advancements in drug development,” but they are not a substitute for regulatory meetings.
“The best way to get [regulators] on board is to present a lot of data that demonstrates what you’re doing works well,” Fisher said. The money Unlearn has raised will go toward continuing its partnering efforts with pharma companies to generate supporting efficacy data for its approach, he said.
Other startups are also in hot pursuit of a cut of the clinical trial enablement market.
Pasadena, CA-based Deep 6 AI raised $17 million in December to move ahead its software, which uses advanced algorithms to process patient data with the aim of finding prospective participants who fit the qualification criteria for certain clinical trials.
Unlearn’s Series A round was led by 8VC, a San Francisco-based venture capital firm that closed a $640 million fund, its second, two years ago. Francisco Gimenez, a principal at 8VC, said in a phone interview that most clinical trial enablement platforms, many of which employ widely available data science tools, aim to make it easier to find the right patients and to streamline enrollment.
But he says Unlearn’s platform could do more—if the company can convince regulators its methodology is sound.
“At the end of the day, we could find all the patients we could want, but … there are a huge amount of human needs and issues with getting someone to join a trial that you can’t AI your way out of,” said Gimenez, who earned a PhD in biomedical informatics at Stanford University. “In a sense, Unlearn virtually found a way to do that by virtue of allowing people to enter these trials specifically to receive treatment or significantly reduce the possibility of receiving placebo, which is a huge barrier, but also, statistically, because we can have these paired experiments, improve the power dramatically. … It’s a rare case of having your cake and eating it too, and so that was really exciting [and] a true differentiator to us.”
As part of the deal, Gimenez joins the Unlearn board of directors. The company’s earlier investors, including DCVC, DCVC Bio, and Mubadala Capital Ventures, also participated in the latest financing.