ARPA‑H is testing a model to make research faster and more interconnected

"It's not one system for everything, but the IGORs would be associated with different diseases, different efforts," Paul Sheehan said.

Interview transcript

Terry Gerton We always love talking to the folks at ARPA-H. You guys are doing some amazing things over there. And today we’re going to talk about IGOR, not Dr. Frankenstein’s lab assistant, but the Intelligent Generator of Research. First of all, what is the problem that ARPA H thinks IGOR can help solve?

Paul Sheehan Thanks for asking. We think that the primary goal of IGOR, the primary problem that it can solve, is really just efficiency and research, right? And so there are challenges to scientists day by day when they’re trying to move things forward, trying to get new therapies for patients, trying to moving forward. And that not only is biology complicated, but their day-to-day capabilities of getting it done can be challenging. And that comes from multiple directions. Some of that is just that there’s so much going on, keeping on track of it, right? And there’s progress being made already in that. But even beyond that, finding what is the next most important problem to solve in an area is challenging. We feel that AI can be an assistant to researchers on that. What’s the next, most interesting thing? And then there are other things, like sometimes I’m doing research and I’m working hard in one area. And then I’m working with AI and it says, hey, I need research all the way in this completely different area. I don’t have that instrument. I don’t have that sound, the freezer, I don’t have that capability, and I don ‘t have a friend down the hall who can do it. How can I go out and find that other person who can do that work? That’s another aspect of this whole program, building out a marketplace. Building out a way to communicate to other people how to get work done. So, if you see like we’re at large, what we’re doing is trying to wrap lots of capabilities around researchers to help them go faster and deliver therapies to patients.

Terry Gerton Is part of the problem that you’re dealing with that we’ve already solved a lot of the easy problems in biomedical research, and now the problems are more complicated, they cross multiple systems, or there’s multiple biological systems, or they cross multiple specialties?

Paul Sheehan Well, as being trained as a physicist, I’m not sure there are any easy problems in biology. They all seem to be complicated. But I do think that we’re at a point where some of the experiments that could have been done to date, you know, they’ve all been done and we’re left with kind of complicated problems where there are lots of different cells doing lots of different things, communicating in many different ways. And that information is inherently difficult for humans to think about. We call it high-dimensional information. That’s one of the reasons AI is so useful because it natively easily works in this high-dimensional space and can identify those problems.

Terry Gerton Walk us through how a research question might actually move through this system once you have it up and running.

Paul Sheehan Oh, great question. Right. So we feel that this process can be applied to many different diseases, many different therapies that one might propose. But if one, you know, were trying to go after a metabolic disease, something like that, right? First, you would use frontier models, the models that are out there that have ingested all the literature to build out like that initial model of how this disease might progress. Then you would use a different tool to analyze that model and go, okay, here’s the gap, right? This is what we don’t know. This is the most important experiment to be done. That researcher can then go, oh, I can do that. And they go off into the lab and get that done. But it could be that that next most important experiment may need to be done by somebody else or a colleague. And then it will help you convey what that experiment is. And this, it doesn’t sound so hard. But it’s one of the main issues facing science, like, what did I do or how should this experiment exactly be run, so you get the same results. That’s part of the program is conveying that information, so that some distant lab can perform that experiment correctly and give you back the information that you’re expecting. It’s a challenge and we’re excited to see what proposals people come up with to answer it.

Terry Gerton I’m speaking with Dr. Paul Sheehan. He is the IGOR program manager at ARPA-H. Dr. Sheehan, you just said something in that description. You said the AI will ingest all of the existing research. Are you confident that AI will draw the right conclusions from all of that existing research? How are you gonna put humans in the loop to say, no, that’s not what would come next. Something else might come next?

Paul Sheehan Well, I mean, that is exactly what we’re going to do. We are putting humans, leveraging all the creativity of humans, their insights, their ability to come up, even new experiments that the AI might not know about so that we can move forward, right? So the AI is going to point to meaningful areas, right, where it thinks the gaps are. And we want absolutely human insight to move things forward.

Terry Gerton The program documents suggest that you could move research forward as fast as 10 times faster than it is today. What would that look like and feel like to the researchers in this space?

Paul Sheehan Well, I hope it’ll feel exhilarating. Everyone, all of these researchers want to solve a problem, right? I mean, you talk to them, they’re passionate. They want to understand what’s going on. They want help patients. So, being able to move even faster, I think is going to be exhilarated for them. And again, we are designing this to solve, to bypass the roadblocks that they run into, right. Keeping up with the literature, identifying the most meaningful problems. Working with others and we’re trying to break down all the barriers so that they can move fast.

Terry Gerton This is not a project yet. It is a request for information and solicitation, I think. How do you envision teams participating in this over the course of a five-year program?

Paul Sheehan So, right, we are in source selection right now. That means we are having a Proposers Day on June 9th for people to come in and team up on those very ambitious program. We’re looking for the initial ideas about a month from now on June 25th so that people can send in ideas. And then it’s really pulling those teams together, identifying the best ones and having them work together. Folks, not only from automation, but the biologists, the researchers themselves, AI specialist all working together to solve this problem.

Terry Gerton You’re imagining teams that are going to cross a lot of boundaries that have been difficult to navigate. How do you hope to see early progress?

Paul Sheehan So this is what ARPA-H does all the time, and this is how we make progress as we pull these disparate teams together to move forward. So part of it is these proposers day, bringing them together. And a lot of computer scientists might come in and is like, I love the problem, love automation, but I don’t know anything about biology or vice versa. And so we put these folks into the same room, they talk about their research and they team up to come up with the best possible solutions for the problem.

Terry Gerton Are there particular problems that you expect folks want to tackle first?

Paul Sheehan So we’re open to a wide range of disease, hoping people take on the most challenging problems. And then beyond that, I do think one of the most challenging aspects is how do we communicate clearly from one person to another on exactly what needs to be done. Communication is always a challenge, especially so in science when you’re asking for something very specific.

Terry Gerton So as this model matures, IGOR, the lab assistant, bringing all of this research together and putting it back out there, what kind of researchers are you expecting to be paying attention so that they — are these university researchers, hospital researchers? Is it international? Who’s gonna be part of the network then that uses the information that IGOR pulls together?

Paul Sheehan So we expect to have the teams using their own IGOR to assemble around different diseases, right? So it’s not one IGOR for all research, but if someone, for instance, wanted to study some rare disease, I would like to stand up an IGOR for this particular disease. It would ingest all the information about it and would direct research in that area, collaborating with the researcher, right? And so everyone sort of would sort of focus on different areas. So it’s not one system for everything, but the IGORs would be associated with different diseases, different efforts.

Terry Gerton So how then do you imagine that it scales does it replicate or does it kind of all grow together over the longer term?

Paul Sheehan Well, we hope, in the long term, it all grows together. Many people are very eager to have a digital twin of humans. Can we just compute everything that’s going on inside the human body all at once? And the way to get there is to build up from component to component. And we feel that’s what IGOR is doing, building up mechanistic models of subparts of humans, and then ultimately, we will piece all these together into a model of the human as a whole.

Copyright © 2026 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.

Related Stories