Personalized Health: Viome's Guru Banavar Discusses Startup’s AI-Driven Approach - Ep. 216

0:10Hello, and welcome to the NVIDIA AI podcast. I'm your host, Noah Kravitz. Health and wellness is one of the most exciting and promising fields being impacted by artificial intelligence today. Researchers and medical professionals have been leveraging AI to improve virtually all aspects of how humans care for ourselves, from advances in medical imaging and drug discovery to helping hospital workers deliver better palliative care. All topics we've explored on previous episodes of the AI podcast, by the way, in case you want to browse the archives for some additional listening. Individualized precision healthcare is another area in which AI is fueling rapid advances.

0:48For as much as we are all alike in many ways, no two people are truly the same, and we all respond differently to diet, exercise, and other factors that influence our health. Viome, a health and wellness startup founded in 2016, is using RNA sequencing technology and AI to focus on nutrigenomics, the science of how food and genes interact and affect the body. Viome's AI-driven gene expression analysis platform focuses on preventing and reversing chronic diseases through deep functional analysis of microbiome and human gene expression. Here to explain Viome's approach and how the progression of AI will continue to shape health and wellness companies going forward is Guru Banavar.

1:32Guru is the founding CTO and head of discovery AI at Viome, where he leads AI and co-leads clinical research. Guru, welcome, and thanks so much for joining the NVIDIA AI podcast. Thank you so much, Noah. It's great to be here. So as is often the case with these intros, there is a lot packed in there, and I'm looking to you to help unpack a little bit for our audience. So why don't we start with the basics? Can you tell us what Viome is, why it got started, and what your approach is to health and wellness? Awesome. Yeah, let's start with that. So let me start with the why question first, right?

2:07So when you step back and think about human health, as you started, there's probably a couple of different categories. There's acute disease for which the healthcare system is built today. And we all know about it. If somebody has a heart attack, you want to rush to the ER, you want to get serviced by the best professionals there are. Then there are infectious diseases like COVID. So you have pathogens which may infect some people, then gets propagated through the whole population. And there's very well-known agents for these infectious diseases. And very well-known solutions as well, like vaccines, for example, and other kinds of mechanisms.

2:48Those are two categories. There is a third category, which I would argue is the most important category right now in the 21st century. And that is the category of chronic disease. Right. Chronic disease or also known as lifestyle diseases are things that you kind of don't really think about a whole lot, but they kind of creep up on you through the course of your life. Right. I mean, you you have your lifestyle, you know, you have a particular way of eating, a particular way of sleeping, a particular way of stressing yourself on being active, not active, whatever it is. And over a period of years, maybe decades, you build up things that you don't even notice.

3:25And until suddenly one day, you know, you have a heart attack or you are told that you have diabetes and you are, you know, worried about, you know, this acute stomach pain. So that is the area that I believe the world needs a better solution. The current health care system is not built for it. the types of knowledge and science that's out there is not great for it. And so what we decided in Viome is to say, let's attack this big issue and make chronic diseases something that we've already solved, just like we solved infectious diseases, we've solved lots of other kinds of healthcare issues. Let's also tackle chronic disease.

4:11That was our main problem that we go after. Now, how do you go after chronic disease? Everybody thinks of genomics or genetics to be a fundamental aspect of your biology, right? I mean, everybody thinks that, okay, if I understand somebody's DNA, I can actually figure out what's going on. Turns out DNA has very little to say about chronic disease, okay? Some DNA variants, like, you know, mutations that may cause, you know, like sickle cell disease or whatever, right? I mean, those kinds of things may over time pick up and catch up and create certain kinds of diseases. But most chronic disease are not based on your static DNA.

4:53Remember your DNA, once you're born until you die, it's more or less the same, right? But your RNA, which is the gene expression that comes out of the DNA is changing very regularly. Depends on your lifestyle, depends on how you slept. you know, for the last few months, it depends on how much activity you have, you know, lots of different things impact, meaning your environment impacts your RNA. So instead of focusing on the DNA, we decided that we would focus on the RNA. Now you might say, why not focus on proteins, which are equally actually downstream of RNA, right? The RNA gets, you know, you know, transcribed into proteins.

5:30Why not focus on protein? Turns out, I mean, we could have done that, but it turns out proteins are way more complicated. You know, your DNA is complicated. RNA is more complicated. Proteins are just super complicated. I mean, in the future, I think we should be doing all of the above. But if I had to just pick one, I think RNA is the place to go for solving chronic disease. So then you ask the question, okay, what difference does it make, right? So let's say that, you know, I knew your gene expression over, not just, you know, over, you know, the last few months, but I knew it over a few years.

6:05Sure. I can actually look at the progression of the gene expression in your body. And I can actually pick up early markers of many things. For example, insulin resistance, right? Insulin resistance comes way before, you know, high sugar, blood sugar. And that is a precursor of diabetes, which as you know, is, you know, 6.5 or more on your HbA1c scale, right? But maybe a decade before that, you can pick up signals of insulin resistance, right? And you can see that in the RNA or you can see that other similar things like inflammatory bowel disease, right? There's all kinds of things that happen when your bowel is in some kind of a relapse.

6:46But when it can be remission for a very long time, it can be in remission for two years and nothing happens, But we can actually pick up what is going on that may, in fact, cause a relapse and, you know, actually prevent it by telling you what might inflame it again. Right. Those kinds of things. So all of those kinds of scientific things can be detected and analyzed and prevented if I were to look at your gene expression. Now, one last piece of this, and I'm going to give it back to you, because I could keep going on for a while on this, is that there is a very significant part of human biology that the medical field has really not taken into account at all.

7:30And that is the microbiome. Turns out that the majority of the cells in our body are coming from microbes. And actually, 99 % of the genes that are in our body are coming from microbes, not from humans. You know, in humans, you have 20, 22 ,000 genes, microbes, you know, you have 2 million genes, you know, so at least, right? And this is expressed genes, right? So there's a huge amount of microbial activity going on in your gut, in your mouth, in your skin, everywhere on your body. And it turns out that is actually a good thing for the most part. But occasionally there's an imbalance that happens because of the community.

8:07And because of the imbalance, you can actually allow certain pathogenic types of organisms to take over. And we actually look at the gene expression. Again, same idea. You don't want to look at the DNA of microbes. You want to look at the RNA of microbes. And you see how that is impacting our immune system, our metabolic system, our circulatory system, all of those things. So we see the interaction between the microbial gene expression and the human gene expression. And once you look at those two things, you can actually understand a lot about chronic disease. So that is why we started. There's lots of implications of all of these things, but I can go to that depending on what you're interested in.

8:47So a couple of kind of baseline questions, I guess. Top of mind, what's a microbe? What's the definition of a microbe? Yeah, a microbe is a single-celled organism. Okay. basically it's a it's you know it's a prokaryote right which is which is that it doesn't have a nucleus it has you know just one cell it has a lot of genes typical you know bacteria bacteria is one type of microbe there are other kinds of microbes like viruses and archaea and fungi and so on so forth but bacteria which are the most common ones that we see in human body they have approximately say 3 000 genes and you know the genes are all over in the cell and you know they have a particular form, but those genes, the 3 ,000 genes are not all bad genes.

9:32Some of the genes, in fact, are very good for you. They may actually be responsible for reducing inflammation because they are, in fact, responsible for manufacturing short-chain fatty acids, which would be very, very important for your gut, for example. So there's parts of the bacterial genome that can do good things for you. And there's parts that are bad things for you, like lipopolysaccharides, for example, are genes that are going to inflame your gut, right? So there are these, you know, there's different types of, you know, the gram negative, gram positive bacteria, and so on and so forth.

10:07There's a whole bunch of, you know, a scientific field microbiology that goes into all of this stuff. But for the sake of this conversation, I think what we need to understand is that these microbes are typically, you know, very tiny, like, you know, single cell organisms, viruses, by the way, they don't even have the coverings. They're just genetic material that are sticking around. They get into potentially a bacteria, you know, like a bacteriophage and can replicate or they can get into a human cell and replicate and so on and so forth. So all of that basic biology or microbiology is important for us to understand before we can get into the bioinformatics and the AI part of what we do in bio, right?

10:46But at the same time, you also want to understand the human biology, right? You know, we are eukaryotic organisms. We have the nucleus, we have genes in, you know, in, I mean, DNA and that gets translated into RNA that gets transcribed into protein proteins, then generate metabolites. That's sort of the fundamental dogma of biology. And what we want to do is sort of intercept it at this point of transcription and say, okay, you know what? You had the same DNA throughout your life, but you know, when you develop diabetes, your gene expression change somehow. Let's catch it before it starts generating downstream proteins and metabolites that may impact your body in the wrong way.

11:24That's kind of what we're trying to do. Got it. And so you mentioned, and not mentioned, it's an important through line of the work you're doing, the difference between looking at RNA as opposed to DNA or RNA as opposed to proteins and the level of complexity involved with both. What's the process of looking at and analyzing RNA and why? Is it just a matter of it's been too hard and too expensive for a long time because we didn't have the technology or why hasn't there been more of a focus? Actually, you put your finger on it. That's always the answer on this show. No, it's actually great. You know, I think it has been very hard to do RNA detection.

12:06And there's a basic reason for it. You know, biology basically generates RNA molecules in a transient way. It's not something that, you know, DNA gets translated into RNA and just RNA stays there for a long time. It doesn't. You know, it can stay there for a few hours to a few days. Okay. So, you know, because of the preservation techniques that we have, you know, we've developed so far, you know, we have not been able to kind of preserve the RNA molecules in their sort of this transient state, that small window of time we have, number one. Number two, the process of getting the informative portion of your RNA versus, there's a lot of non-informative portion of your RNA.

12:48You know, there's a ribosomal RNA, which is very sort of housekeeping oriented. There's a lot, there's like 90 something percent of your RNA is just these housekeeping, you know, transcripts, which are not really useful. Right. I mean, not really useful in understanding chronic disease. So if you take the, all of the RNA, let's say you preserved it and you took all of the RNA that's in any given sample, you may end up spending a lot of money just processing all of this useless, you know, non-informative RNA. So you want to sort of figure out how to get the informative portion of the RNA. And then even then, you know, you go to some place, you know, you go to an Illumina sequencer and you run a batch, you know, it's called a flow cell, right?

13:33You run a batch of samples through an Illumina sequencer. There's a lot of contamination that happens. If you have, let's say, 100 samples on a batch, when you look at a particular sample, it turns out that 1 % to 5 % of the sample are coming from the other samples on the same batch. So there's a crosstalk that happens, which is taken as a given in the industry. Even Illumina accepts that as a given. But that's not good enough. You need to get, if you're doing machine learning, you don't want to have noise that is above 1 % where signal is actually below 1%. So what you want to do is you want to bring the noise down to the floor of the noise down to a level where the signal to noise ratio is reasonable enough that you can pick it up and do some analysis.

14:21So all of these are problems that the RNA community has not been able to solve for a long time. It turns out our team in Viome, we started, you know, one of our, you know, we had like three major streams that came through in Viome. One of the streams was the RNA processing, you know, technology. Our chief science officer developed that in the Los Alamos National Labs for a different application domain. And we were able to kind of, you know, get that, you know, patent license to Viome. And that already had already built all the solutions to these problems that I just mentioned. so we got high quality data right and then once you get the high quality data the second stream was my stream which was the ai stream and machine learning what do you do with it how do you understand it and then how do you turn them into clinical applications that second stream and the third stream is sort of the you know how do you commercialize all of this stuff you know there's many ways of doing it you can go direct to consumer you can go through professionals many different ways and there's business model things that we can talk about as well sure that's a third person from our founding team, the three of us put our heads together and sort of started Viome and said, this is how we're going to solve this problem.

15:30Right. Gotcha. And so Viome is direct to consumer now? Well, we started direct to consumer. Okay. And for the first four or five years, we've been only direct to consumer. But just very recently, just since last year, we've started doing professional solutions as well. And so we have oral health solution, oral health pro, and in the future we plan to do gut health pro and so on and so forth. So we go to professionals as well, but we have a very solid consumer, direct to consumer business and for good reason, I think. Yep. So let's walk through it then from the consumer perspective, if that's right, or consumer patient, whatever.

16:06But from the perspective of somebody like me, I sign up, so to speak, with Viome. Yep. What happens? I get a testing kit? Yes, yes. So, you know, think of us as an e-commerce business, right? So you go to our website and you select one of the multiple different types of testing kits we have. Okay. And, you know, we have gut health, which is gut intelligence, which only requires a stool sample. Okay. And then we have our premium kit is called full body intelligence, which is what I would recommend to anyone who is listening to this podcast. Okay. Try it. it's full body because you know at the end of the day when you're when you're looking at the human biology that everything is interconnected and you you want to understand what's going on in all parts of your of your body so we started with the digestive you know system right you want to look at the top of the digestive system which is which is your saliva you know just a non-invasive saliva sample okay look at the bottom of the digestive system which is your your stool you look at what's going on in the gut.

17:07And then we also look at around the digestive system, which is your blood, you know, which is picking up nutrients from the digestive system. So all three samples, saliva, blood, and stool is part of the full body intelligence kit. And, you know, if you buy that, you get a kit with all three kit, all three sample tubes, and you collect those samples, send them back to Viome, you know, prepaid envelope, all of that stuff. And then you get your results on an app and you get your molecular pathway insights, like your gut scores, your oral health, your immune health, your brain health, your kidney health, your, you know, the whole body.

17:48We can actually tell a lot of things from the three samples. Okay. Just through those three samples. Got it. You can, then you have the option. Actually, you end up getting recommendations about what kinds of foods are compatible with your biology. right there could be a lot of problems you know you could have too much uric acid you could have a lot of inflammation you could have you know a bunch of you know different things going on in your body and we can say listen you don't want to eat foods that can you know that can uh have a lot of oxalates if your oxalate processing oxalate metabolism is not doing well i don't know what's an oxalate oxalate is a molecule that uh you know that that ends up getting metabolized in in your gut, you know, actually a lot of green leafy vegetables like spinach and many others have a lot of oxalate in them.

18:43So if they're not metabolized, if they're metabolized, then it's great. And microbes, in fact, are a key part of metabolizing them. But if they're not metabolized, they coagulate and they become kidney stones. They don't get discreeted. So if your oxalate metabolism pathways are not active and we can detect that just by looking at you know the transcripts that are that are in your system so if they're not active then you don't want to be eating you know even though Popeye told you that's right it's healthy right you you don't want to be eating too much spinach at least for a period of time right and maybe you can reinvigorate your oxalate pathways by eating you know different types of foods like you know fibers of certain types of prebiotics and activate them.

19:32And then once the oxidative pathways are back up and running, then you can start eating more spinach because then you can metabolize the oxidative. Right. Okay. So you get back nutritional recommendations. You get back nutritional recommendations. And then we also identify the superfoods for you, things that would be awesome for you to eat. So we say, you're going to metabolize these things. You're going to, you know, you're going to reduce your inflammation, whatever it is, you know. So we say all of the superfoods for you. And then we give you another option. We say, listen, if you don't want to modify your diet dramatically in order to get your superfoods, we can actually take a concentrated version of your superfoods, put them in a customized supplement pack for you.

20:19Okay. And send that to you, you know, with your name on it. We send you like, you know, 30 packets every month. So you can actually consume that instead of, you know, or in addition to, I should say, consuming your superfoods to the extent you can, you make sure that you get all those nutrients that your superfoods are providing for you. Got it. In these supplements, prebiotics, probiotics, lozenges, and very soon we're going to have a probiotic and prebiotic toothpaste as well, which actually takes care of your oral health. Oh, for oral health. Okay. Yeah, oral health. So all of those things are possible.

20:52as a customer slash consumer of Viome products. Now, on the professional side, you go to your doctor and the doctor may say, hey, look, you look like you have a lot of white patches in your mouth. You know, I want to just check this out. And so it turns out we were able to detect a molecular biomarker for oral and throat cancer using the same technology. Oh, wow. Okay. And we got an FDA breakthrough designation for that a couple of years ago. And we turned that into a professional product. So you go to the dentist and your dentist may say, hey, look, you know, I don't like what I'm seeing. I'm not sure what it is.

21:33Let's just get it tested. So you spit, you know, in a tube and you get a report that says, hey, you know, is the biomarker for oral throat cancer, even early phase, you know, at least stage one found. because typically, unfortunately, oral and throat cancer is found in the stage three or stage four, at which time your mortality is very, very high. So, you know, like more than 50%. So you can reduce that by early detection. There's a lot of things you can do to remove that. So if you went to your doctor, you might get one of these kinds of tests. And that's also supported by the same platform that Viome has.

22:12And not to neglect the thousands of questions going through my head, but for the sake of the conversation and to get into talking a bit about how you're using machine learning and AI, I'll narrow it down to this. Okay. What is your success rate, to put it that way? And you can interpret that in the best way possible. How confident are you in the analysis and the predictions coming back when you analyze these different samples? Yeah. So I think the gold standard in the medical industry is randomized clinical trials. Right. Right. So take, you know, the usual methodology is you take a set of people who have a particular condition, you put half of them, let's say, in a placebo arm, and you put the other half of them randomly in an intervention arm.

23:06And the intervention in this case is this analysis that we've just gone through, right? You know, you look at all the samples, you look at all the pathways, you figure out what the right ingredients are that this person has to be taking, what are the things that they should be avoiding? That's also part of it. Sure. And then have them follow that regime for a certain period of time, let's say three months, and then ask the question, what is the difference between the intervention group and the control group over that period of time? So that's the gold standard, right? So we have been running a few RCTs and we just got some intermediate results.

23:38we see that diabetes goes down more significantly in the intervention arm than in the control arm. Okay. So HbA1c drops by, I think, 0.45 % or something like this. You can go from diabetic to non-diabetic, meaning pre-diabetic, or if you were pre-diabetic, you'd go to non-diabetic if you follow this process compared to a controller, right? Similarly, we also did this for depression, major depressive disorder, which is the endpoint there is not HPA1C. It is a clinically validated endpoint called PHQ-9. It's a questionnaire where you score somebody on a scale of 1 to 27. And you ask the question, what is the initial?

24:22The initial point is it's got to be moderate or severe. And then you ask the question after three months, what is it? And again, there we are seeing the same kind of impact. Right. Oh, that's fascinating. So we've done those RCTs and we're just starting to publish those. But before that, we measure the biological pathways themselves. And we can tell when, you know, you look at 10 ,000 people or 100 ,000 people and you see what the scores for any given, like inflammation, for example, right? In the gut, you can see what the distribution of scores are. And you can say, okay, if your inflammation is super high, it's not a good thing.

24:55You know, it's got, you know, if it's super low, it's great. And in the middle, you can do better if you did the following things, right? So we also have done studies in which we see, we asked the question of, did somebody's inflammation score go down over a period of, let's say, six months? And we published those results. We've measured glycemic response. You know, we took 1 ,100 people and we asked the question, you know, we gave them food. We gave them specific foods for a period of two weeks. And we had a continuous glucose monitor. and we said, let's look at all of the responses, glycemic responses for every single type of food over actually more than a year in this particular case, both in the US and in Japan.

25:35We did two cohorts. And then we found that we could create a machine learning model that given a new food and an individual's gut microbiome and saliva microbiome, we can predict which foods are going to create a high glycemic response for you versus for me. and you know for you banana may be uh you know high glycemic response for me bread may be high glycemic response yeah and the reverse you know for you for you bread may be low and for me banana may be low so if i knew that you know i would reduce one and increase the other because then i can manage it right we've done all of those kinds of things already and we have something like i don't know 15 or so publications right now and every couple of months we put out i mean you know we put out new peer-reviewed publications and you know it's almost like you know we are we are i I mean, I have so much data we could be publishing every week if I had the time to write it down.

26:25Yeah. So we have a ton of data and we are always thinking about what is the best way to to publish in peer reviewed journals and make sure that the community sees and evaluates our results and tells us what is OK and what is not OK. Right. What is a good result? What is not a good result? So this is part of our scientific sort of foundation for VIO. Right. So let's transition a little bit into the AI side of things. The company was co-founded in 2016, so about seven or so years ago. And you mentioned kind of the three strands of the company, figuring out the RNA part of it, business model part of it, and then sort of in the middle of it, your domain, the AI part of it.

27:11Yeah. How has I don't know what the best way to ask this question is, but how has the use of machine learning evolved over the time at since Viome was founded? And maybe talk a little bit about as much as you can anyway, about what the platform does, what types of analysis it's really good at and tuned to, what kinds of things maybe it just hasn't been so good at. And then a little later, we can get into perhaps what's coming down the pipeline. Sounds great. So first of all, the data that we have in Viome is important for everyone to keep in mind, right? So a sample, right? Any sample. Let's say you send us a saliva sample.

27:55Yep. We can detect every single microbe that is known to man, basically. Roughly how many are we talking about? So our catalog right now is in the range of, I mean, if you think about organisms, we think, I would say it's in the range of about 40, 50 ,000 genomes that we already know are fully assembled and annotated. Yeah. Right. But actually, what is more important from my perspective is not the genomes, it's the genes. And our catalog for the microbial genes is in the range of 100 million right now. Okay. 100 million. So that's a catalog, right? So your sample comes in, we sequence your sample, and that turns into a whole set of what are known as sequencing reads, right?