AI is moving beyond text, images, and code. Now it’s learning to read and write — shall we say program — DNA.
In this episode of NEXT, I chat with Eric Nguyen, co-founder and CEO of Radical Numerics, about the rapidly emerging world of biological AI. Nguyen and his team helped create Evo and Evo 2, generative foundation models for DNA, and are now working toward what they call “general biological intelligence” … AI systems capable of understanding biology across DNA, gene expression, methylation, proteins, and other biological signals.
The potential upside is enormous.
These systems could help scientists detect cancer earlier, develop treatments for antibiotic-resistant superbugs, understand disease more deeply, and eventually design countermeasures to emerging biological threats on demand.
But the same capabilities introduce serious risks.
Nguyen explains how AI could potentially generate biological sequences that retain dangerous functions while evading traditional sequence-matching detection systems … essentially creating biological “deepfakes.” He also discusses why AI labs need to develop biodefense capabilities alongside increasingly powerful biological design tools.
The conversation covers:
- DNA foundation models
- AI-generated viruses
- biosecurity
- pathogen detection
- wastewater surveillance
- attribution of biological threats
- antimicrobial resistance
- cancer detection
- open-source versus closed-source biological AI
- why biology may be the next major frontier for artificial intelligence
Transcript: AI can now edit DNA and create deepfake viruses
John Koetsier: Well, if you weren’t worried enough about AI escaping containment, hacking third-party systems, and getting access to things it shouldn’t get access to, check out OpenAI and Hugging Face if you haven’t heard about that yet. Now AI is editing DNA. And there’s some amazing things there. We know that we can heal some diseases by editing our DNA. There’s some challenges as well, too. Deepfake viruses that are not digital but biological. We’re going to get into all of it with our guest. Eric, say hi and introduce yourself.
Eric Nguyen: Hey, John, good to be here. My name is Eric. I’m co-founder and CEO of Radical Numerics. We are an AI research lab, and what we make possible is for AI to be able to read, write, and program function into DNA. The idea is broadly to train language models in tree-like language to be able to train AI and understand this raw substrate of life directly.
John Koetsier: That makes some level of innate sense, right? I mean, DNA is a language. It encodes information. It tells an organism how to develop, right? So you’re learning that language and being able to express that language, and probably maybe the first one to use AI to do that.
Eric Nguyen: Yeah, so we started in this field about three or four years ago, and our team is known for creating these models called Evo and Evo 2. And so, up until that time, 2024, folks had tried to use AI to be able to read DNA. DNA has been historically extremely challenging to understand. Actually, most of our DNA, most of the genome, we still don’t understand what it actually does. Parts of it we do, and we’ve leveraged that to make medicines and understand disease, but there’s still so much we don’t understand. In fact, people call it the dark matter genome because it’s everywhere, but we can’t quite see it and know what it does.
And so the hope was that AI can help us look into and interpret what’s in our DNA. And that’s been, for a long time, one of the major goals. And what we tried to do along the way was not just be able to read DNA, but to write it, and so to be able to generate it. In the same way we generate natural language, and seeing how that really just blew up and changed the field of natural language and chatbots and artificial intelligence broadly, we wanted to bring that kind of innovation in frontier technology to DNA, which is one of the most complex, if not the most complex, language of all.
John Koetsier: Super interesting. I mean, I think it was just a month or a couple months ago that we saw scientists create an artificial cell. And there’s debate over what that means, what that looks like, whether that’s living or not living or anything like that. But now you’re actually being able to write DNA, potentially inject that into a cell. We’ve all seen this movie, Eric. How does it end?
Eric Nguyen: Yeah, I don’t think anyone knows what the end means, but what I will say is that I’m much more of an optimist than pessimist overall. The promise to be able to harness and control the function and program the function into something like DNA has far more benefits and potential impact on human health. This is the ultimate, in my mind, the ultimate benefit of AI. And so I think ultimately we’re going to have a far better understanding of how disease works and live far longer and healthier with this type of technology growing.
That being said, I think it does open up biological risks, meaning if you lower the barrier, make it easier to design things for positive benefit, there’s also the potential of misuse, as we’ve seen with every type of technology. And so, as a company, one of the things that we built into our mission from the foundation is this dual mandate. So we apply this frontier technology to push on the biological design capabilities, as well as actively use it for defense, meaning we deploy this and make sure that it can be safely used and actively deploy it to make sure that it’s not in the wrong hands, essentially, of folks that want to misuse it.
John Koetsier: Yeah, I mean, Pandora’s box in some sense, right? I mean, the continuum from amazing tool and what can it do? So many diseases have a genetic component or can be solved potentially with editing genes or applying a new gene-specific vaccine or medication. That’s absolutely incredible. Talk about that side first. What do you hope to do? What do you hope to crack first? What do you hope to fix first?
Eric Nguyen: Yeah, so I think, stepping back about why we thought this was such an interesting problem, there’s obviously a lot of excitement about the potential of AI to work in drug discovery and human health. I think the focus has been on what I call single-modality problems, or sort of like the low-hanging fruit, meaning usually the way you understand and pursue things in biology or disease is you really narrowly focus on one type of, let’s say, molecule, like a protein—
John Koetsier: Mm-hmm.
Eric Nguyen: —or an RNA molecule, or even one disease at a time. And I think the bottleneck is going to be much more complex as we figure out these single-modality challenges, because we can do things like design, you know, a protein structure with things like AlphaFold. And the bottleneck is going to shift from, can we make that specific one molecule, to how is that thing going to behave and react in a body as something complex—
John Koetsier: Because we can’t do things like all approach each other.
Eric Nguyen: —and inherently multimodal, right? So a system as complex as your cells and your tissues is not just one molecule. It’s a cascade of events that happen inside a living organism. And so what we haven’t seen is technology that’s able to model such complexity because we’ve so much specialized in the past. And so the technology we’re building is much more of a general-purpose engine to model biology. We actually call it a general biological intelligence.
And so what we want to be able to do is not just design the molecule or the drug, but to model how it’s going to behave in a system. And that’s the innovation that we’re opening up.
John Koetsier: Unbelievably interesting. I mean, I’m going off memory here. I’m thinking we have something like a trillion cells in our bodies. Does that sound about right?
Eric Nguyen: Yeah. Yeah, yeah.
John Koetsier: Chaos theory comes into play here, right? I mean, everything is overdetermined, everything has multiple causes, multiple effects, everything affects everything else. How on earth are you going to model this level of complexity? Are you going to eventually use quantum computers for this? And how are you going to do it?
Eric Nguyen: Yeah, and I think it’s a great, great question. And one of the things that folks in the domain of science and biology like to bring up is that data is so important, and I agree. Data is extremely important, like any space in computation, especially AI. The thinking is usually that there’s not enough data in biology, and I actually think the opposite. I think that there’s too much data in the space.
The bigger challenge is we don’t know how to use it. We don’t actually know how to model and essentially be able to generalize across different settings and biological contexts. And so the anecdote I like to bring up here is that just for DNA, for example, DNA sequences, there’s more DNA online than all the text on the internet. So we’re talking internet-scale information, and that’s just for DNA, and that’s just one of the modalities—
John Koetsier: Yeah.
Eric Nguyen: —and there’s dozens of other modalities. And so there’s massive amounts of data in the space. And what you see other tools and models basically be able to do with this data is have something that’s basically as powerful as running linear regression or using an Excel spreadsheet. It’s just this big gap in terms of how much data there is and how much signal we’re able to actually extract from it. That’s been the challenge.
I think this is what we’re excited about as an AI research lab. We are building models that can actually learn how to scale and glean from this massive amount of biological data and to be able to not just ingest it, but to actually be able to infuse it and have these inherently multimodal models. That’s, in my mind, our best chance to be able to model something as complex as biology.
John Koetsier: Super interesting. Is all that DNA information freely available? Open source? Is that somebody that’s donated it? How is all that DNA information available?
Eric Nguyen: Yeah, so just one website, it’s called the SRA, or Sequence Read Archive, is basically a government or nonprofit entity that collects DNA sequences that are sequenced around the world and just hosts them. And so this is, you can think of it like the national archive of genomes that folks can just download.
John Koetsier: Okay, wow, interesting. Okay, so there’s huge opportunity here, and we’ve just started to scratch the surface of what that looks like, what that could make available. There is a downside, there is a potential downside. If I can go on some LLM and say, “Make a virus,” that’s pretty freaky. And I’m not saying that we’re there yet or that we’ll ever get there.
But reality is that for every lab that’s working in the light, there might be a lab that’s working in the shadow, right? And we don’t know what’s going on. That’s the argument on the AI side as well: we have to keep building because the bad guys are building too, right? And so there’s a strong argument that you have to build for the defense side. What do you mean by a deepfake virus in biology?
Eric Nguyen: Yeah, so deepfakes in general, right? People usually may associate this with images, the idea that AI can generate things that look very similar to, let’s say, natural images. For viruses, what we mean there, obviously it’s a general description, is the idea that one could design DNA to essentially function like a virus, but be able to obfuscate or intentionally basically switch the letters around—
John Koetsier: Or intentionally based on analysis, is the left.
Eric Nguyen: —so that existing detection systems cannot actually notice that it’s a virus that they’ve seen before. So it’s rearranging the alphabet or rearranging the letters in a way that maintains the function, but then does not match what’s been seen before.
And there’s a couple of things to back up and set the context for why this is such an important and critical thing. One that I think a lot of folks don’t realize is that it’s possible to essentially send in a design of DNA to these DNA synthesis companies. And what they’ll basically do is physically synthesize it for you and send it through the mail, and you basically get it through the mail like an Amazon package. And so this runs the biological research community, and so it’s used for good. It powers the research to understand disease and making drugs.
But then the concern is that if folks try to make something dangerous like a biological weapon, there should be a minimum level of screening and protection so that someone doesn’t try to order the bubonic plague, for example. And so there’s this list of not-allowed dangerous sequences to be generated. And so if you send something and it matches the spelling to a database of dangerous things, then it’s crossed off the list. You cannot do that. Clear enough, simple enough, everybody agrees with that.
Now, what AI now allows you to do, and this is something that we showcased a few weeks ago when we announced a preview of our next-generation models, is that these AI biological foundation models, which are trained to be able to generate new sequences of proteins and DNA, you can then actually intentionally have them design things that can get around these detection systems, right? And that’s a concern that’s a growing concern for a lot of folks, especially at the national security level.
You can imagine, for example, if a country has a history of using biological weapons or has a program, this just adds one more sophisticated technique to get around another country’s defenses for this, right?
John Koetsier: And presumably, if you designed a flu virus with something that was just changed a little bit, then perhaps the body’s immune system wouldn’t recognize it and wouldn’t fight back against it. And you could have something that is otherwise, hmm, I’m down for a couple days, to fatal.
Eric Nguyen: Yeah, I think this part gets especially scary, right? The idea of not just getting around detection methods, but then intentionally making things more pathogenic or things that can cause disease and more harmful, well, is especially scary when you think about how good AI is when you give it a certain goal, right? It’s just able to optimize through many cycles. And this just overall brings the, you know, the way I categorize this is, just brings the barrier for making more and more dangerous—
John Koetsier: Mm-hmm.
Eric Nguyen: —things, right? As you make more capable design tools, it goes hand in hand with allowing or enabling biological risks like this to be used for potential harm. And so that—
John Koetsier: Mm-hmm.
Eric Nguyen: —is why we felt it was so critical for an AI research lab to really make this an inseparable mission, to drive both and to make sure that the tools that you’re building, you’re cognizant of what you’re enabling for harm. And at the same time, if you’re going to work on this safety element and biological defense side, you need to be at the frontier of the capabilities, right? Otherwise, you won’t know how to defend against these things that are emerging.
John Koetsier: Yeah. Mm-hmm.
Eric Nguyen: And so—
John Koetsier: Mm-hmm.
Eric Nguyen: —when we look to other companies and other folks in the space, we just saw that gap. You usually saw that was siloed. The entities that worked on the frontier research worked on only the design capabilities. Entities that worked on the safety side only worked on the safety, but they didn’t really understand the tech. And so this is why we felt it was so critical for a company like ours to really make this core to our mission.
John Koetsier: How do you put technology like yours that you’re developing and will come out with shortly, whether that’s months, years, whatever it is, how do you put that on the front lines such that as doctors find, get aware of, are exposed to a new disease that looks like something but has a different morphology, different action, different level of lethality, whatever it might be, how do you put your technology on the front line so that you can quickly detect that, understand that, and work on some countermechanism?
Eric Nguyen: Yeah, great question. So I think one of the more pleasant surprises when we first started thinking about biodefense is that, from a detection standpoint, or that service of being able to detect something’s dangerous, it’s kind of naturally an API-driven business, meaning you have a sequence that you want to characterize. And so send in a sequence and get back a score that characterizes and just says, you can basically think, dangerous or not dangerous.
You can have some more information about where does it come from, what is it most similar to, which parts of the sequence are especially dangerous, have dangerous pathogenic proteins. So from that standpoint, we are especially compelled by the idea of making a centralized system where folks can just upload sequences that they’ve collected from the environment, whether it’s a hospital, emergency room, or there’s programs that actually surveil wastewater treatment plants, like sewers, that can detect from a population—
John Koetsier: Mm-hmm. Mm-hmm. Mm-hmm.
Eric Nguyen: —level if something is anomalous, right? Is there something floating around? Or using it as an early detection system for potential attack. Yeah, yeah.
John Koetsier: Hint, hint, there is.
Eric Nguyen: And so that’s a piece. And then, for governments and the national scale, there’s definitely interest in terms of having frontline accessibility of detection technology. So being able to know if your soldiers on the battlefront come across especially infectious bacteria, you want to know that fast, but also, as you mentioned, what do you do about it once you have that knowledge? So, worth talking about—
John Koetsier: Mm-hmm.
Eric Nguyen: —in my mind, the three pillars of biodefense. So the first is detection, right? Surveillance and detection, basically being able to find it first. The second is attribution. So it’s not enough to just know if it’s dangerous or not, but especially for governments, you want to know where it came from. You want to know who is it attributed to. Is it natural? Is it man-made? Is it an attack? Does this come from a specific lab in a certain country, for example? That’s especially top of mind given the most recent pandemic.
And the third is countermeasures. So what can you do once you do identify the threat? And so there’s this area of being able to rapidly design countermeasures that folks are especially interested in in this age of AI. Can you make an antimicrobial or antibiotic quickly for a superbug, for example, especially in hospitals? Can you make antivirals? Can you make antitoxins?
These things, the search space, or I should say the possibility of possible pathogens, is so vast that this old regime of trying to stockpile all the potential, you know, anti-whatever, is not practical. And so what becomes more practical and looks more sustainable is having an AI system that can rapidly design on demand. And I think that’s one of the things we’re especially excited about and hopefully can showcase in the near term that we’re not quite yet talking about, but absolutely interested in.
John Koetsier: Wow. And so every hospital, every healthcare center has their own machine where they can, on demand, synthesize the treatment for a particular disease. That would be amazing.
Eric Nguyen: That would be incredible. And it may not have to be as distributed as that. It could be a little more centralized so it can be a little bit faster. But yeah, there is a world where you can synthesize on the spot as well. I think what’s more common is that you screen on the spot, or technology is called sequencing. So you put in the biological sample into, like, a Petri dish, and then it would read out the actual DNA. And then the challenge then becomes, okay, how do I read—
John Koetsier: Mm-hmm.
Eric Nguyen: —and interpret this DNA? And that’s where the AI systems can do this rapidly.
John Koetsier: How far away are we from this future, Eric, where we need to be aware that these things are possible? And whether it’s nation-states, whether it’s some cult, whether it’s just some crazed billionaire who’s in a movie fantasy, whatever the case might be, how far are we from a future in which governments need to be maintaining always-on programs, scanning for these things and instantly ready with a rapid response?
Eric Nguyen: Yeah. So I’d say now, and I think this is a concern for frontier labs especially. I think what you’re seeing—
John Koetsier: Because nothing like that is ready now, is it?
Eric Nguyen: So there are existing programs that do have screening and surveillance capabilities in the U.S., for example, and other countries. The challenge has been that they’ve been severely underfunded for a very long time. And so—
John Koetsier: Mm-hmm. And sort of—
Eric Nguyen: —for better or worse, the progress of AI has made folks really assess different scenarios that are the most threatening and dangerous, right? And if you—
John Koetsier: Different scenarios that are the most threatening if you have a—
Eric Nguyen: —when you look at what Anthropic and OpenAI and DeepMind, for example, put out in terms of the response to what are some of the worst-case scenarios, biological threats from AI is often the number one concern. And so I agree in the sense that—
John Koetsier: Friend of the—
Eric Nguyen: —this is something of a concern now and only going to grow, in my opinion. Because it’s one of the reasons why we initially got into this direction of defense. We sort of played out the scenario in a world, for example, where—
John Koetsier: Mm-hmm.
Eric Nguyen: —you have agents, billions or trillions of them, running around the internet making and taking all sorts of actions. People have a concern about cybersecurity, and they absolutely should be. But there’s also folks that make biological tools like us, obviously our intentions are good, but in the hands of these agents, there are disappointingly very few tools on the defense side to make sure that what they’re creating is actually safe.
So that was a big reason why we felt like we need to even out this sort of arms race between the design capabilities and the defense capabilities. Right now, the defense capabilities are just far, far behind, and we need to bring that back up.
And so I think that’s the sense of urgency that folks are, I think, starting to wake up to now. There’s more movement we’re hearing from Capitol Hill about putting forth bills together about a more strategic and holistic strategy against biological threats.
John Koetsier: That was not a comforting answer.
Yeah.
Eric Nguyen: Yeah, yeah, yeah. And so I think it’s not meant to purely scare, but I think it is trying to bring a dose of reality and urgency. At the same time, I do think that there’s an opportunity and a lot of interest emerging from the frontier labs, from folks like us. And I think we all have our different roles. As, for example, I think frontier labs are definitely doing it from restricting their access to chatbots in certain ways. And whenever you bring up the word virus, for example, it pretty much shuts down, of course, or in this case.
But I think the limitations that we see there that we’re trying to work on, and that we think we absolutely do have solutions for, is that AI can understand some of these threats from these chatbots from a natural-language standpoint. They can understand the series of questions that one asks, or try to understand if someone is trying to design something dangerous, but they never can actually understand the sequences themselves. They can never get past reasoning on the raw biological data because it hasn’t been trained in a way that can actually reason across it. It’s trained on language that then describes those things.
And so this is what we feel in terms of capabilities. We train our language models directly on the biological sequences so that we can reason across and find the motifs and patterns that cause something to be actually harmful or not. And so this is why we think, and what we showcased a few weeks ago with our model called Omni, that the ability to detect these deepfake viruses, it is possible. And we have comments from folks that would tell us in the biosecurity community that they did not know it was possible to do this, to detect this kind of threat.
And so I’m an optimist in terms of making sure that this technology is used for both good and against bad. And what we also want to do is make sure that folks in the AI research community are aware that this is a challenge and to build up a research community of folks excited to work on these problems. Because I think the age of text and video and audio is sort of like the low-hanging fruit, and I think the AI battles on those modalities and on chatbots, people have sort of figured out those recipes, how to build in that space, and they’re going at coding agents.
But the next frontier, in my mind, is this space: the physical world, biology, and science. And in this sense, people do not know the recipes. And so there’s a huge opportunity for impact here.
John Koetsier: I do want to get to some of the positives in a moment. I want to talk about cancer. I want to talk about some of the other major diseases that we face right now and the potential of this technology to address those. Before we get there, the history of AI has been largely one of collegiality, working together. And I’m talking historical time over decades, right? And this is what we’re working on. Share it, you know, this is what we’re working on. And we see that today with models that are released, open-sourced, sometimes even with model weights and other things like that.
You release your models as well, open source, which makes some level of sense because your data is open source, essentially, as well, that you’re training them on. However, when you release these models open source, anybody can grab them, anybody can use them. What’s your rationality for releasing these things that could potentially be used for very dangerous purposes openly?
Eric Nguyen: Yeah, great question.
John Koetsier: Yeah, great question.
Eric Nguyen: When we worked on Evo and Evo 2, these are the first generative DNA foundation models and the largest biofoundation models ever, that was a top concern. And so one of the things we did to safeguard against the potential misuse was to remove pathogenic sequences, specifically viruses that can infect humans, so removing that from the training set.
The thinking, the hope, was that making it this much more difficult for folks to be able to use it for nefarious purposes. And we made the calculation overall that because of where the field was at for DNA foundation models, very early, right, very early, the potential good and forward movement on the research side was going to far outweigh the negative that could potentially be used from these models.
I think that calculus has started to shift since then. And so I think overall I would say it depends on the context and the time. I think in one moment in time, the right decisions may seem like open source, but yeah, if you’ve seen—
John Koetsier: And so that’s what we’re doing.
Eric Nguyen: —things like mythos and these other capabilities, when it gets to a certain level of capabilities, then you kind of reassess, right? And so I think we’re seeing something similar like that on the biological foundation models. And that’s why our next generation of models, Omni, they are initially closed source, and especially because they are trained on these dangerous sequences. And so then we control access for the defensive purposes to our allies and to partners that are using it for societal benefit, essentially.
So the potential for harm was especially worth bringing up in consideration when last year folks had used our model Evo to generate the first genome from scratch. This was published online last year by a group. And what they created was what’s known as a bacteriophage. And it’s also known as a virus. So they did create a virus with our models, which at face value sounds terrifying to folks, but under the hood, it is a virus that can only attack bacteria. It cannot physically attack humans, so it’s, in other words, harmless to humans.
John Koetsier: Mm-hmm. Fortunately, we don’t need any bacteria in our gut at all.
Eric Nguyen: Yeah. So, yeah, the idea is that they wanted to be able to train it to target specific bad bacteria, so infections.
John Koetsier: Gotcha.
Eric Nguyen: So that was the motivation. And in fact, that is one of the things that we work on as a company, is to train these models to generate bacteriophages essentially as an antimicrobial. Because the other part, the other flip side of this, is there’s a whole host of folks that contract some kind of infection that is not responsive to antibiotics. Sometimes people call them superbugs, and actually there’s two million folks a year that die from this, right? So there’s a lot of people that still are infected and succumb to these bacterial infections.
And what one can theoretically do, and people have done this with natural phage, is select these phages and basically use them as an antibiotic and target—
John Koetsier: Mm-hmm. Mm-hmm. Mm-hmm.
Eric Nguyen: —a specific strain of bacteria. So not kill all the bacteria, like keep the good ones, but target the bad ones. It’s possible. It’s done in low doses, done in low volume in terms of people. And so that is one of the examples where there’s a natural benefit, but then there’s also this sort of natural fear of what happens if it does jump and turn into something dangerous to a human. And so these are the things that folks just inherently have to balance when they are working at this intersection of health and bio.
John Koetsier: Intersection of health and bio. What a challenging thing. What a challenging thing. You have the entire complexity of inventing and evolving this rapidly changing and expanding field of study, and you have the added complexity that if you do the wrong thing, bad people could get access to powerful tools.
And even if you do the right thing but you make a mistake, or there’s just something you’re not aware of, or who knows what, these things can happen as well. This is an extremely challenging area, and I don’t think we’ve—I don’t think we have any clue what it’s going to look like in 20 years. What power capabilities we’ll have, and just hopefully our defensive capabilities evolve just a step ahead of the attackers.
Of course, that never happens, right? You always get attacked, you understand, then you develop a defense. You get attacked, you understand, then you create a defense. So I hope that we get better and better at rapid detection and early intervention.
We did talk a little bit about the positives there, and you talked about superbugs and being able to treat people like that. And I’m sure that that’s a growing thing as we have more and more antibiotics in the world and in our livestock and everything, and superbugs arising because of that. There are potentially other areas as well. Cancer is a massive killer, obviously, and there are many other diseases as well. Talk briefly, if you would, about the potential for this technology to help on those fronts.
Absolutely.
Eric Nguyen: Absolutely. And this is my favorite part. This is what motivates us to work on this technology. One of the things that we’re especially excited about is cancer detection, right? So there’s lots of research and literature that shows that detection is one of the best things one can do to basically make cancer survivable. The earlier you can—
John Koetsier: Mm-hmm.
Eric Nguyen: —detect, the more options one has and the less it’s progressed.
The challenge has been that the earlier that you try to detect, the less signal, the less basically evidence that it’s actually in your body. And so being able to detect some of these very weak signals kind of comes down to finding the typos, right? The typos, if you were to use that DNA analogy, or that language analogy in DNA, finding that thing that caused mutation that not just mutated, but actually leads to cancer. Not every mutation leads to cancer.
And then also, it’s not just DNA, but there’s a host of other biological molecules that give these signs of cancer. It can come from your blood. It can come from little changes in these chemical binding properties to your DNA. It’s called methylation specifically. It’s basically some kind of—
John Koetsier: Mm-hmm.
Eric Nguyen: —chemical change that changes the shape. DNA actually is kind of bundled up into a 3D shape inside your body, inside your cells. These little changes can cause different cascades of events that are just really hard to pick up traditionally. And so traditionally, folks have usually tried to detect these typos, I guess, one molecule or one signal at a time.
And it’s kind of changed throughout the years. It’s like, we found this molecule that, if we read it and we find the typo, it’s going to be easy to detect. And so it’s kind of been a promise in cancer detection over the years of this is the holy grail. In our mind, it’s not just one thing that comes along. In our mind, it’s the idea of fusing all of the molecules and being able to read all of them to be able to detect these very faint signals.
And so my analogy that may be easier for folks to help visualize: if you’re trying to model or, let’s say, GPS, when you’re trying to find your phone, you’d have these satellites that kind of try to triangulate and have different ways to kind of—
John Koetsier: You know, trying to—
Eric Nguyen: —grab the signal. And what people do now in biology, it’s kind of like trying to use one satellite at a time and optimize that one satellite. And instead, what you want to do—
John Koetsier: A lot of the—
Eric Nguyen: —is you want to use all the satellites to triangulate, right? And so the technology that we’re building—
John Koetsier: We recognize that to be able to discover—
Eric Nguyen: —concretely will fuse DNA, will fuse gene expression, will fuse methylation, all of these streams of information that one can basically sample and test from your body. We are giving it all to the AI to be able to then decipher across all of these signals, all of these sensors, so to speak.
John Koetsier: A lot of these acceptors, so speaking, as well as satisfying scanning process, that’s all of these really differently.
Eric Nguyen: And so some people like the word sensor, or phrase sensor fusion, to give this idea of how do we merge the signals across all of these very different, typically distinct specialties. And that’s one of the reasons why we thought the field has been held back when folks have really tried to specialize—
John Koetsier: A lot of these acceptors, so speaking, as well as satisfying scanning process, that’s all of these really differently.
Eric Nguyen: —one thing at a time. This systems approach, this holistic approach, is what we think is going to power things like cancer detection, to make it far more able to basically triangulate on the signal of cancer. And that’s one example, and I think there’s many across other diseases as well.
John Koetsier: I can’t tell you how exciting that is, because I probably can’t, because you’re very excited about it yourself, but I’m excited about it as well because we’ve siloed things so much in medicine that you get this test for that, this test for that, you get a doctor for this, you get a doctor for this, and getting a holistic picture seems to be harder and harder as we’ve segmented everything. That can only be super helpful, whether it’s in cancer or other diseases.
Eric, super interesting stuff. I wish you a lot of wisdom as you proceed on these challenging but very, very promising endeavors.
Eric Nguyen: Thank you so much. It’s been a pleasure, and I’m very excited, and I hope folks are excited about this as well.