[Adobe stock]
But the company’s pedigree is unique. The firm was founded as a generative AI research company in 2017 before genAI was cool. “We were saying generative AI was going to be the future of AI research,” said CEO Charles Fisher.
The three co-founders of the company hold Ph.D.s. “All three of us co-founders have backgrounds in physics, ML and AI,” said Fisher, who earned his doctorate in biophysics, focused on protein folding. “We used physics-based models for that kind of work. This was before any of this deep learning stuff happened at all,” he said, referring to the rise of AI-powered tools like AlphaFold, AlphaFold 2 and Meta’s ESMFold. Another co-founder, Jon Walsh, head of modeling, has a doctorate in high-energy particle physics. Aaron Smith, the third co-founder and machine learning scientist, is a Ph.D. mathematical physicist. “Jon and I tried to read [Aaron’s] Ph.D. thesis once and we only got through the second sentence,” Fisher said.
GenAI models to model clinical trial participants
Charles Fisher, Ph.D.
The founding thesis for Unlearn centered around using generative models capable of learning complex simulations directly from data. “When we think about building a digital twin of a person, we want to have a computer model that takes in data from that person and allows us to forecast what can happen to them in a variety of future scenarios,” Fisher said.
The idea of developing mechanistic models to simulate biological complex systems remains elusive. With tens of trillions of cells, the complexity of the human body makes it more feasible to use some “kind of an abstraction,” Fisher said. “We simplify it by saying we’re not going to try to understand or directly model the mechanisms, chemistry or genetics. We’re building deep generative models to take in data and forecast things in the future. That’s what we mean by a digital twin.”
Cutting enrollment needs
Unlearn’s focus now is to use digital twins, in this case, digital patient models that forecast a patient’s future health using generative AI models trained on extensive patient-level data from previous studies, to rethink clinical trial design. Its technology could substantially reduce patient enrollment requirements. While the specifics depend on sponsors’ needs, the technology aims to reduce the need for clinical trial enrollment between 25% and 50%. “If you have a clinical trial with 1000 patients and you can remove 25% of the control group, that could translate into four to five months of shorter time to complete enrollment in your study,” Fisher said. “And 50% is major. That can actually save you close to a year in terms of your clinical trial timeline.”
Unlearn’s updated TwinRCT can now incorporate additional historical clinical trial data to strengthen its power for Phase 2 trial evaluations. A recent internal analysis of a completed Alzheimer’s disease trial showed that the company’s TwinRCT 3.0 offering could achieve the same statistical power as an analysis of variance (ANOVA) analysis with 23% fewer participants, resulting in an estimated reduction of five months in the enrollment process.
The concept of using digital twins in clinical trials has won support from some Big Pharma firms. An article in Expert Opinion on Drug Discovery notes that Roche, AstraZeneca and GSK are exploring applying digital twins in the clinical trial process. GSK has also explored digital twins in vaccine production.
Gaining regulatory support for digital twins in clinical trials
Unlearn’s digital twins are created by training AI models on vast amounts of patient data from past clinical trials and real-world sources. “We’ve aggregated data from thousands of clinical trials and observational studies that covers about a million patients,” Fisher explained. “For a dataset of this type, that’s a very large, high-quality clinical trial and observational study dataset. It covers something like 30 different indications.”
The digital twins can then predict how a patient will respond to a treatment, allowing researchers to reduce the size of the control group while maintaining the same statistical power as a larger trial.
Unlearn’s technology has already gained regulatory recognition. The company has gone through qualification with the European Medicines Agency for using this approach for primary analysis of phase 2 and 3 studies. “We have recently got a letter from the FDA basically saying they concur with EMA’s qualification,” Fisher noted.
Expansion and future plans
Unlearn is now looking to expand beyond its initial focus. “We initially started in neurology, working on Alzheimer’s, ALS, Huntington’s and things like that. Now we’re starting to expand into new indications in immunology, metabolic disease and other areas,” Fisher said.
The company recently raised a $50 million Series C funding round to support this expansion and invest in further research and development. “We’re going to put more resources into our R&D efforts. We really are an AI research company and our goal is not just to build something for today, but to invent new technologies that will allow us to solve the problems of the future,” Fisher explained. “So we’re also putting significant efforts into R&D, thinking about what our next generation technology should look like.”
Filed Under: clinical trials, Data science, Drug Discovery, machine learning and AI, Regulatory affairs
