How AI can unlock the mysteries of our health, with Kevin Slavin

KEVIN SLAVIN: There are these major shifts in consciousness over time. For a long time, we believed that the earth was the center of the universe and then a bunch of science gets done and that goes away.

And then, we think that our species is the center of all life. And then a bunch of research gets done and that’s obviously false. There’s this third one that I think is still in progress, which is the notion that each one of us is an individual organism. I’m trying to help get us past that fallacy as quickly as possible. Because we’re not, because we’re filled with tens of thousands different organisms in our gut. Those have profound effects on our outcomes and who we are.

RANA EL KALIOUBY: That’s Kevin Slavin. He is a dynamic thinker in the field of metagenomics, which is how we learn about microscopic organisms like bacteria. These microbes live around us, on us, and inside of us.

SLAVIN: I think we generally consider air and the environment in general as a white space, right? Like empty. And of course that’s absurd the minute you start poking at it. It’s incredibly rich in genuinely millions of species that have direct outcomes for all of us, not just pathogens, also lots of stuff that just straight up keeps us alive and, in many cases, helps us thrive. And all of which are invisible to the eye.

EL KALIOUBY: Each of these tiny microbes have their own characteristics. And figuring out what is what, takes a LOT of compute. Here is where AI comes in.

AI can detect patterns in the DNA of these microbes far quicker and more efficiently than we ever have been able to before. This is a big deal. Because, like Kevin said, these tiny species, that we never see, have real impacts on our bodies and our world.

Which is why on this episode we are digging into the significance of metagenomics, our gut biome, as well as our environmental biome. Once you understand how microbes shape us, you’ll understand why it’s so ground-breaking that AI can help decode them.

I’m Rana el Kaliouby and this is Pioneers of AI – a podcast taking you behind the scenes of the AI revolution.

[THEME MUSIC]

Hi, Kevin. Welcome to the show.

SLAVIN: Thank you.

Copy LinkWhy microbes force us to rethink what it means to be human

EL KALIOUBY: So there are a bazillion directions we can take our conversation today. But I wanted us to particularly focus on this idea. A lot of your work, especially recently, has been focused on understanding and managing the microorganisms and the microbes both around us and inside us.

So for some of our listeners who are not familiar with this space, can you give us a sense of the complexity and the vastness of both the number of microorganisms that are kind of within our body, but also around us?

SLAVIN: So, these are estimates, right? But we understand there to be about six and a half million species of life on land and maybe like another 2 million in the sea. So call it eight and a half million species on earth. We have names.

And at least a crude understanding of about 2 million of those, right? So the space of everything that’s left is enormous. The part of what was interesting to me when I first started this is how vast this world is and how little we know about it, and how would we even begin to interact with it in order to understand it? And that’s on and off what I’ve been working on for about 10 years.

Copy LinkHow metagenomics makes the invisible world measurable

EL KALIOUBY: So how do you do that? So now we’ve got 2 million of these that we can at least name. What did your work do?

SLAVIN: What I was interested in initially, and what has become a very urgent and widespread field, is can we find out what’s around us — all these organisms that we can’t see that we might not even be able to name — how would we even detect them?

There are different aspects of this kind of investigation. The first thing is that you need to be able to collect the DNA that you’re going to be looking at. So you have to be able to gather all the genomic material, then you need to be able to figure out what you’ve got. Part of what was making all of this possible right around that time was kind of the advent, and lowering costs, of what’s called metagenomics.

And here I will pause and explain what metagenomics is because it’s just not part of most people’s breakfast. So if we were to talk about genomics I would say okay, well, you spit into a jar and I know that you’re human. And I look that up against the fully characterized human genome and I can make a bunch of conclusions based on the specific variations within your genome.

Okay, but if I give you a bowl of soup, what do you look it up against? Do you look it up against the tomato genome? Do you look it up against the cow genome? Do you look it up against pepper genomes? Like, where would you even begin? And you can’t, right, you have no idea where to begin, especially if you really want to find out everything that’s in the soup.

So you have to do what’s called metagenomics. And metagenomics is basically saying, okay, look, we have no idea what’s in the soup, so we’re just going to break up all the DNA of everything in there. And we’re going to look that up against the entire tree of life, which, from a computational perspective, is ambitious.

And so that’s quite a task. Back when we were first doing it in 2014, it took many days and was really displacing a lot of computational resources. These days it’s a lot better and a lot easier.

Copy LinkWhy AI is essential for decoding complex microbial DNA

EL KALIOUBY: Probably because there’s a lot of harnessing machine learning.

SLAVIN: Exactly. Part of it is just computational strength, right? There’s GPUs that specialize in this kind of thing and so on and so forth. But really it’s the ability to use machine learning and AI to start to do some baseline predictive aspects.

EL KALIOUBY: Kevin is talking about what to do with the genomic information once you collect it.

But there’s still the question of how to even collect this data. If you’re analyzing someone’s gut biome, you analyze their fecal matter, basically their poop! But what if you want to measure the environmental biome of Boston. How do you collect data on that scale?

Back in 2014, Kevin found an inventive way to do just that.

Copy LinkCreative ways to collect genomic data from the environment

SLAVIN: I arrived at using honeybees to gather the DNA for us. It’s what honeybees do. You don’t have to pay them, although they do need help to survive. And they are optimized by millions of years to basically gather.

Every microbe that they come across, pollen is the most obvious one, but also everything else that they fly through or touch, adheres to their skin. They’re basically little swabs, that are broadly sentient, and can go and gather.

And so we built hives that would give them something to wipe their feet on when they got home. And then we would send whatever the bees had found to the Weill Cornell Center for Computational Genomics, and they would sequence whatever we had gathered, and that sequencing is the second step.

But here I should say that back in 2014, I was using honeybees, but through that work I also met and ultimately became friends with and collaborators with folks who were using other techniques. So, one was Jessica Green, and she was using HVAC filters.

So if you think about the filter in your air conditioner or your heating system, it’s purifying your air. But from a microbial perspective, that’s like a fishing net. And it catches everything. And so she would collect things through the HVAC. And there was Chris Mason, who at the Weill Cornell Center for Computational Genomics in New York, who was using grad students at that time. They went into every subway station in New York City and physically swabbed every single subway station to find out what was in the subway station.

So you can use bees and you can use HVAC filters and you can use grad students. And then there was a grad student, named Marianna Matus, who was in the Eric Alm lab. And she built, back then, the very first robot to go into the sewers and collect essentially wastewater, to find out what was in there.

EL KALIOUBY: A huge fan of Mariana, by the way. Shout out, Mariana.

SLAVIN: Shout out to Mariana. Truly a genuine pioneer. But I think with all of us, we were all fascinated by this capability to gather genetic material and genomic material from the environment.

And really at that time everyone just sort of thought that was a bit of a folly.

EL KALIOUBY: A folly because back in 2014, the world didn’t know why we would even need to collect and analyze the genomic material of a city.

That of course changed by 2020, when COVID upended our lives. The need to understand which invisible microbes were in our environments suddenly became critical.

We’ll get to that in a minute after a short break. Stay with us.

[AD BREAK]

Copy LinkHow pandemic surveillance turned environmental sampling into a public health tool

Take us to like January or February 2020 and how this ability to both collect and then understand what’s in the air around us. How did that become really important?

SLAVIN: I would say that in March of 2020, there still weren’t diagnostic tests available. And yet here I was talking to scientists that I’ve been talking to for years saying, I don’t know how to find it in somebody’s body because we didn’t have rapid antigen tests yet or even PCR, but I think that we all know how to figure out if it’s in the room. We could do that, couldn’t we? And so for a while, the very first instantiation of this — I would say by March of 2020 — was what’s called surface testing, which is basically using the same swabs that go up your nose, but swabbing your coffee cup, the doorknobs, the elevator buttons in your building, and so on and so forth, and looking those up against what was now a fully characterized SARS CoV 2 genome.

We put together a program for the World Trade Center in New York mid 2020 so that they could get a handle on when COVID was present. This is a complicated way to do things.

It’s fraught. You’re not detecting it really in the environment, you’re really detecting it on, Lisa’s coffee cup.

So that was not destined for long term success. And I moved on to working with some folks at another company that had figured out how to do airborne detection of COVID. And that was using electrostatics. It would basically pull air through a fan, effectively, and then bind microbial material to a metal plate.

And then we could take the plate, and sequence whatever we found on the plate. And like that, you could find out not whether somebody in particular had it, but whether it was present in the room — was somebody exhaling it? And so there was a moment, a few months, where detecting COVID in an office or a hospital or whatnot was a useful signal. But then with the advent of Omicron, once that happened, it was a little bit like saying, I’m going to sell smoke detectors right outside a forest fire. It was a piece of information that nobody needed. Is SARS CoV 2 present in the Starbucks?

Yes. Is it on this bus? Yes. What about this school? Yeah, right. So that was also not really a viable business. And so I migrated over to Ginkgo Bioworks, which is a biotech company out of Cambridge, and they have a division now known as Ginkgo Biosecurity that was sort of taking the same questions and sort of scaling them up. And the ways that they were scaling them up was to say, okay, it’s not useful to know that it’s on somebody’s coffee cup and it may not be useful to know that it’s in the school.

But from a national security as well as a public health perspective, it may be useful to know what just entered the country. It may be useful to know what just entered a secure space of some kind. And in particular, if you can get full genomic information about what you’re capturing.

So in other words, not just saying SARS CoV 2, what leads to COVID, is present, but also which particular variant is present. So really being able to capture the full genomic profile of the things that you capture is half the problem, right? But also, what do you collect to give you the significant signal?

And in the case of Ginkgo Biosecurity, they entered into a program with the CDC called Traveler Genomic Surveillance, TGS, which in, I believe the number is 10 airports.

In those airports when airplanes land, the 250 passengers get off the flight, and then the little truck comes up to drain the wastewater from the plane.

And instead of just sending that more or less to a landfill, a sample of that is extracted and brought to our lab. And that could tell the folks at Ginkgo Biosecurity not just who landed, but also what landed with them.

And to take it one step further, it’s not just who landed and not just what they brought with them, but also where did they bring it from.

And that’s where you get into a kind of data information that is especially interesting if you work in the intelligence community or national security, because you’re learning what is happening in other countries. It’s a way to gather a genomic landscape at a global level without having to deploy specific genomic surveillance in every location.

Copy LinkUsing genomic intelligence to track biological threats and origins

EL KALIOUBY: So this is really powerful because it enables you to determine whether a virus or a biological agent is kind of originating from nature or it’s created in a lab.

Because this distinction is important, right? Because it has, obviously, national security and biosecurity implications.

SLAVIN: Absolutely. There are three general threats from a biological perspective to, let’s just call it, civilization or national security. Those three threats are things that are natural. It used to be that we would say that there was a 1 percent chance every year of something that was kind of COVID scale. And I would say that most epidemiologists and other folks in the space would agree that it’s generally around 3 percent risk every year, due to things like climate change and human encroachment on animal habitats, things like that. And 3 percent every year adds up quickly.

You do that over 20 years and it’s a coin flip of whether we have another consequential flu or COVID-like event. So that’s one type of threat. Another threat is lab leak, which may also be natural, but were not intended to be consumed outside of a lab.

And then the third risk is something that is genetically engineered and adversarial. And that is a thing that keeps the folks who know a lot about it up at night. And then the question is, if and when somebody uses something like that, how would we know? Not Ginkgo Biosecurity, but Ginkgo Bioworks, developed a tool called NDAR, which is basically a means by which you can look at a sequence of DNA and evaluate whether this was made by nature or by humans. Now if you think about the computational task there.

You’re looking at something that has never existed before, and then the question is: does this look like something nature did or something that humans did? And that is the kind of thing that AI is really necessary for and really good at, because what it needs to be able to do is to look at the entire space of all nature in the world, all genomic information in the world, as well as the history of all genomic information in the world.

Copy LinkWhy healthier buildings should be designed with microbes in mind

EL KALIOUBY: I am fascinated by this idea of building microbe-friendly architectures and urban design. Tell us more about that. And how do we bring that kind of thinking to our schools and homes and hospitals?

SLAVIN: Yeah, I grew up in New York City in the seventies. Most of the buildings in the Lower East Side were all empty. Why were they empty? They were empty because in the 20s there was a series of epidemics in New York, including tuberculosis, that were so serious and so prevalent that the city just said, okay.

The pathogen is definitely a problem, but the other problem is that people are living in such cramped conditions in environments that don’t have circulating air, that don’t have sunlight. And so they made a law that said every room in any place that you live in New York City needs to have sunlight and air. And at that time in New York City, the landlords who were not wealthy people themselves couldn’t afford to make those changes to the buildings, and so they just kind of walked away. But it was illegal for anyone to live in them because they were so dangerous from a pathogens prevalence perspective. We designed our water systems around cholera. We have always designed and redesigned the ways that we live around the biological threats that we have faced, until COVID, where we mostly just shrugged our shoulders. It’s a horrifying missed opportunity.

EL KALIOUBY: There’s so much that the microbes around us can say about our health and safety. And while there has been a missed opportunity since COVID to redesign our indoor spaces, I’m still hopeful. AI will accelerate our understanding of the pathogens and all other kinds of microbes in our environments.

But there’s a whole other microbic world living inside of us. We get information about them through the waste our body produces, and there’s a lot we can learn by understanding them. AI can help.

That’s after a short break.

[AD BREAK]

Copy LinkWhat the gut microbiome reveals about health risk and disease

All right. Let’s switch to the microbes within us and how important these are, because these microorganisms aren’t just hanging out in our gut. They’re actually really key to our energy production, nutrient production, our health and wellness. So tell us more.

SLAVIN: Well, we’re at the phase of knowledge production where we know that they are deeply consequential. And we know a little bit about some specific ones that matter, but we are still really untangling what does what. We need to pay more attention to it. The state of the human microbiome, which is very different than it was a hundred years ago, is consequential to our outcomes, and that could be everything from our immune system to autoimmune diseases, to who knows what. The first one I came across that felt very sort of concrete was just, there’s a specific microbe, Toxoplasma gondii, that is in one out of three.

Americans. And if you have it in you, you don’t know that you have it. It’s not, it doesn’t announce itself, but it will change your risk profile. You are far more likely to crash your car or become an entrepreneur or all the other things that involve poor assessments of risk. And so is that you, or is that Toxoplasma gondii? It’s basically the interaction between you and something that you can’t see.

Each one of us is teeming with life and it’s not neutral. It has very profound effects. There’s something called C. diff, C. difficile, you can pick it up in particular when you’re in hospital and your immune system is suppressed. And C. diff doesn’t necessarily kill you, but it’s a very aggressive disease and it’s wholly resistant to antibiotics.

And the most effective treatment for it is a fecal matter transplant. There was a case — this is about 10 years ago. These were some of the earliest fecal matter transplants for C. diff, right? So they used, in quotes, healthy poop to repopulate his gut. Having nothing to do with the C. diff, he also had alopecia. He had lost all of his hair.

And there are lots of people who have alopecia, and it’s just a thing that happens, it’s just a mystery. So here’s a person who had alopecia and C. diff, and they used healthy poop to address the C. diff, and all of a sudden their hair started growing back. We still don’t really know what alopecia is, but now we know something about how it is produced, and it has something to do with something in your gut.

EL KALIOUBY: This is where AI can play a really huge transformative role, right? Because the relationships between these microorganisms and the things we ingest and our biologies, it’s so complex. And this is where AI can hopefully help.

SLAVIN: I agree.

EL KALIOUBY: Out.

SLAVIN: I agree. The computational space is so vast. And I think where AI can help is essentially being able to just notice patterns that would be otherwise invisible. And those patterns allow you to engage these data sets with far better predictive power.

Copy LinkThe promise and limits of personalized microbiome medicine

EL KALIOUBY: One of the areas that I’m really interested in is, once you’re able to really understand a person’s microbiome, and also like skin biome, right, oral biome. Can you then use that to create personalized medicine, right? And maybe that’s for me, like personalized probiotics.

This is exactly the mix of things that I need to take. What needs to happen for us?

SLAVIN: I think at a systemic level, it feels a bit far at the moment.

EL KALIOUBY: Because we don’t know enough. Because we don’t know.

SLAVIN: Yeah, we don’t know enough. And because our interaction with the world, which changes our microbiome to some degree, is broadly stable in many respects, but it’s not fully stabilized ever. And so then, the question is like, is the thing that you’re taking this month relevant to what you should take next month? Will it be relevant if you move to a different part of the state or if you travel to another country? Probably not. I think the idea of personalized medicine is still built around the idea of a discrete individual. And here we’re talking about a kind of collection of organisms that’s in flux and will remain in flux, so that is a vexing problem.

EL KALIOUBY: To me, it also comes back to, absolutely yes. And what kind of sensing technology, like, how are we collecting that data? Right. Because if you are getting a stool test every three months or a blood test every six months, it’s not dynamic enough.

Like we’re not getting enough sample points.

SLAVIN: Yeah. And really stool — poop — is really the only, it’s the single best way to find that out. And yes, there are smart toilets that’ll, those exist. They will continue to exist. They will get smarter.

EL KALIOUBY: Is not the Japanese type of smart toilet that can change temperature and stuff. This is actually a toilet that can analyze your.

SLAVIN: Yeah. I would call those other ones kind of wise toilets. Those are like toilets that can help your mood. But smart toilets basically can do the kinds of essentially genomic surveillance on your own stool.

It’s certainly possible to do that for things that are targeted, like if you’re looking for specific things, which include cancer. But the idea of being able to pick up things that you weren’t looking for is still a serious computational task and the types of extraction that need to happen for that won’t be built into your toilet this year.

EL KALIOUBY: Well, if any of our listeners are working on these smart toilets, like, please do reach out. We’d love to learn more.

SLAVIN: Actually, to one of your earlier questions about how pathogens have shaped the environment, it’s a fundamental dynamic that the overall, in particular in the United States but globally, effort to just reduce the number of microbes to which we are exposed has done what it wanted it to do. But it also did a bunch of things that we didn’t want it to do, which is basically reduce the diversity of the microbes in our guts and our skin and our mouth. We need better tools to be able to evaluate why that matters. It may well be right that if we’ve eliminated our exposure to any number of microbes, maybe that’s okay, except for the following seven.

But we don’t know yet. And that is really the edge of the edge.

And if we are ever able to answer that question, it will be because AI has become powerful enough to.

EL KALIOUBY: Yeah. Amazing.

Copy LinkA human vision for AI that preserves wisdom across generations

Okay. Last question. If you could have AI do anything in the world for you, what would you have it do?

SLAVIN: Wow. That’s a good question.

If I could have AI do anything for me, I would want something actually very specific. I have thought about this, which is that I have a nine year old daughter.

And someday I’ll be gone. I don’t want to live forever. And I don’t have much interest in the folks that do.

But I really like the idea of leaving something behind so that she can continue to talk to me. Not pretending that I’m alive, but I lost my father very early. I often wish that I could ask him things.

EL KALIOUBY: What would you, what would you do in this situation or what do you think of this or—

SLAVIN: Yeah, or how did you handle it?

I want something that can scrape everything I’ve said, everything I’ve written, everything I’ve done and be able to deliver that, in a conversational form, to my family forever, when I’m not around to do that myself.

EL KALIOUBY: That was a powerful answer.

Kevin, thank you for coming on the show. That was so fascinating.

SLAVIN: Oh, thanks to you, Rana. I love your show and the wisdom that you bring to the domain. I really, really appreciate it.

EL KALIOUBY: AI will revolutionize every facet of our lives, especially when it comes to our health and wellness.

For example, how can AI personalize our vitamin intake? I take about 15 vitamins and supplements every day, as do a lot of people I know.

I think it’s helping me … but I would love to know the exact impact all these supplements are having on my body. And even better, I would love to get a personalized regimen … And this is where AI can help! By the way, if you’re building something like this – please do reach out!