Terry Gerton We always enjoy chatting with the team at ARPA-H. You’re doing incredible work there. Today, we’re discussing IGOR—not Dr. Frankenstein’s assistant, but the Intelligent Generator of Research. To start, what issue does ARPA-H believe IGOR can address?
Paul Sheehan Thanks for the question. We believe IGOR’s main goal, and the core problem it can tackle, is truly about making research more efficient. Scientists face daily hurdles as they try to advance their work, develop new treatments, and push forward. It’s not just that biology is complex, but the practical, day-to-day process of conducting research is also demanding. This complexity comes from several angles. Part of it is simply managing the sheer volume of information and staying organized, though progress is being made there. Beyond that, figuring out the next critical question to answer in a field is tough. We see AI as a valuable partner for researchers here, helping pinpoint the most promising next step. Additionally, a researcher might be deeply focused on one area, and an AI tool might suggest a crucial experiment in a completely different field. They might lack the specific equipment, expertise, or a nearby colleague to perform it. How do they find the right person for that task? That’s another key part of this initiative: creating a marketplace, a platform for connecting researchers and outsourcing work. So, in essence, we’re aiming to surround scientists with a suite of tools to accelerate their progress and bring therapies to patients faster.
Terry Gerton Is part of the challenge that the simpler problems in biomedical research have largely been solved, leaving us with more intricate issues that span multiple biological systems or scientific disciplines?
Paul Sheehan Well, as a physicist by training, I’m not sure biology has any truly easy problems; they all seem complex. However, I do believe we’ve reached a point where many of the straightforward experiments have been completed. We’re now left with highly complex problems involving numerous cell types performing various functions and communicating in countless ways. This kind of information is naturally difficult for the human brain to process. We refer to it as high-dimensional data. This is precisely why AI is so powerful—it inherently operates well within these complex, multi-dimensional spaces and can help identify key issues.
Terry Gerton Could you walk us through how a research question might flow through this system once it’s operational?
Paul Sheehan Excellent question. We believe this approach can be applied to a wide range of diseases and potential therapies. Let’s say, for example, you’re targeting a metabolic disease. First, you’d use advanced AI models that have processed all existing scientific literature to build an initial model of how the disease might develop. Next, you’d employ a different analytical tool to examine that model and identify the knowledge gap—the most critical unanswered question. The researcher could then tackle that experiment themselves. However, it’s possible that the next vital experiment requires a different specialist or collaborator. The system would help clearly communicate the details of that experiment. This might sound simple, but it’s a major hurdle in science: ensuring an experiment is described precisely enough that another lab, anywhere in the world, can replicate it and produce consistent results. A core part of our program is developing ways to convey this information accurately, so a distant lab can successfully perform the experiment and return the expected data. It’s a significant challenge, and we’re eager to see the innovative solutions proposers suggest.
Terry Gerton I’m speaking with Dr. Paul Sheehan, the IGOR program manager at ARPA-H. Dr. Sheehan, you mentioned that the AI will process all existing research. How confident are you that AI will derive the correct conclusions from that vast body of work? How will you incorporate human oversight to correct potential missteps?
Paul Sheehan That is precisely our plan. We are integrating human expertise—leveraging human creativity, insight, and the ability to conceive of novel experiments the AI might not consider—to drive progress. The AI will highlight significant areas where it detects knowledge gaps. We absolutely want human judgment and insight to guide the next steps.
Terry Gerton The program documentation suggests research could progress up to ten times faster than current rates. What would that accelerated pace feel like for researchers in the field?
Paul Sheehan I believe it would be incredibly exciting. Every researcher we speak with is passionate—they want to solve problems, understand mechanisms, and help patients. The ability to work even faster would be thrilling for them. Our goal is to design systems that remove the common obstacles they encounter: staying current with literature, identifying the most impactful questions, and collaborating effectively. We’re striving to eliminate all barriers so they can move swiftly.
Terry Gerton This isn’t a formal project yet; it’s a request for information and solicitation. How do you envision teams engaging with this over a five-year program?
Paul Sheehan That’s correct. We are currently in the source selection phase. This means we’re hosting a Proposers Day on June 9th for interested parties to form teams for this ambitious initiative. We’re expecting initial concept submissions about a month later, on June 25th. The process will then involve assembling these teams, selecting the strongest proposals, and having them collaborate. We anticipate teams comprising not just automation experts, but also biologists, researchers, and AI specialists—all working together to solve this challenge.
Terry Gerton You’re envisioning teams that will bridge many traditionally difficult boundaries. What kind of early progress are you hoping to see?
Paul Sheehan This is what ARPA-H specializes in—it’s how we drive progress. We’re looking for early demonstrations that these integrated teams can effectively combine their expertise to tackle complex, multi-disciplinary problems. Success would mean showing that the collaborative framework and AI tools are genuinely accelerating the research cycle, from hypothesis generation to experimental validation, across different labs and specialties.
Here is the paraphrased version:
We bring these diverse teams together to drive progress. That’s partly what our proposers’ day is for — gathering everyone in one place. Many computer scientists might show up and say, “I’m passionate about the problem and love automation, but I have no background in biology,” or the other way around. So we put these people in the same room, they share their research, and they form teams to develop the best possible solutions.
Terry Gerton Are there specific problems you anticipate people will want to tackle first?
Paul Sheehan We’re open to a broad spectrum of diseases, and we hope people will take on the toughest challenges. Beyond that, I think one of the biggest hurdles is how clearly we can communicate exactly what needs to be done from one person to another. Communication is always difficult, especially in science when you’re asking for something very precise.
Terry Gerton As this model evolves — with IGOR, the lab assistant, gathering all this research and sharing it back out — what kinds of researchers do you expect to be paying attention? Are these university researchers, hospital researchers? Is it international? Who will be part of the network that uses the information IGOR compiles?
Paul Sheehan We expect teams to use their own IGOR to organize around specific diseases. It’s not one single IGOR for all research. For example, if someone wanted to study a rare disease, I’d like to set up an IGOR dedicated to that particular disease. It would absorb all available information about it and guide research in that area, working alongside the researcher. Everyone would focus on different areas. So it’s not one system for everything — each IGOR would be tied to different diseases and different initiatives.
Terry Gerton So how do you envision it scaling — does it replicate, or does it all grow together over time?
Paul Sheehan In the long run, we hope it all grows together. Many people are very excited about the idea of a digital twin of a human being. Can we compute everything happening inside the human body all at once? The path to getting there is to build it up piece by piece. That’s what we believe IGOR is doing — constructing mechanistic models of smaller parts of the human body, and eventually, we’ll connect all of these into a complete model of the whole human.
