Latest / Two PhDs on a Pod / From Poop to Data: Understanding the Language of Microbes
Transcript
- speaker-0: you Hey, mama. speaker-1: ⁓ so again, we're talking about microbes, of course. So there are two microbes in the colon of some dude, and they're bragging about their beneficial functions for the dude. And one says, Hey man, I produce a lot of butyrate. I must be the host's favorite microbe. And the other one says, what are you talking about? I make more than you. I must be his favorite. And then a third microbe comes in and says, Whoa, whoa, whoa. You two are acting like the host knows you. All he cares about is your poop. speaker-0: Poop! Okay, the metabolites? speaker-1: Yeah, metabolites, yeah. So I'll explain this a little bit for those people who are like, wait a minute, what are we talking about? There's poop and there's poop. See, poop is what people think of human poop, but there's also the micro poop, ⁓ is what this joke is about. So when we talk about microbial metabolites like butyrate, they're literally the poop of the microbes. Microbes take the fiber that we feed them. and they convert it to butyrate to harvest the energy from those fibers and they literally poop out butyrate. Now the fact that we have evolved to depend on that poop to live and to have a proper physiology, that's a different story. But yeah, so what I call that is micro poop. speaker-0: Sounds cool. And you know, I love the fact that we have Normalized the word poop. I think you poop is an important part of life, know, it's no absolutely human poop the microbial poop and everything else so I think Our audience is gonna love this episode because we're gonna talk about how we turn poop meaning the human poop Into data that can actually make a difference for you and your health. But before we go into that I have a pretty big announcement to make. Who? Yeah, I was just recently officially inducted as an honorary member of Gen Z. What do think of that? speaker-1: Officially? Well, sounds official. Who's the official? speaker-0: was an entire committee of Gen Z people who voted on this and they said, okay, we'll end up this guy. I had to work hard for it, mama. It was not easy. There was a lot of things. I'm going to talk about it a little bit later in the episode. So if you want to hear about that, you need to stick around. when I heard your joke, I said that first microbe was just aura farming. that's a Gen Z. Wow. And then I want to do one more Gen Z thing, is we're to play Dera 10, but and the question I'm going to ask you is the microbe that works tirelessly in my gut. What is the rating of that? speaker-1: That is a 10. speaker-0: But the fact that the greatest medical contribution to my life from that microbe is technically its poop. What is that? That's a six seven, I think. speaker-1: Okay. Okay. Definitely. Talking to a Gen Z here. Exactly. Wow, Guru. Congratulations. That's a big one. Awesome. Well, I'm Momo. speaker-0: Thank you, man. And I'm Guru. We're two PhDs on a FOP. speaker-1: All right. Let's dive into it. So what's on your radar, Guru? speaker-0: something. Okay, well, this is a good tie into what I was going to talk about with Gen Z and everything. Over the weekend, my daughter, Livy, graduated from her neuroscience degree at the University of Pittsburgh. So let me actually show you a beautiful picture of the building that I was super impressed with. It's called the Cathedral of Learning. speaker-1: I really need to make it to Pittsburgh. It's one of the few remaining places in the US that I want to see and I have not. speaker-0: And when you go there, you should definitely check out this cathedral of learning also known as Addy. So Livy standing in front of the cathedral of learning. It's, you know, it's a beautiful 1787 building like long time ago and it's built in the Gothic style and it's a very tall building like 42 floors or something. So you can actually go all the way up. And you know what happened was we were up in that building and we were sitting in one of the classrooms, which by the way are all modeled. for different countries in the world. Super, super cool. And Livy started talking to us about some of the things that she had learned in her neuroscience course in her degree. And it turned into an intro to neuroscience. And there were so many cool things she talked about. Like, for example, there's this myelin sheath that is covering the dendrite and helps the transmission of signals and messages from synapse or from a neuron to another neuron across the synapse synaptic junction and so forth. It's such an important function because otherwise if you did not have these myelin sheets, it would leak essentially the electrical signal that needs to pass through. And it makes it much more efficient to transmit those signals. So I'm sure we're going to talk more about that, Momo, in the future about the importance of that for health. I think you had some important observation to make about that. What was that again? speaker-1: Well, yeah, I mean, it's just a little teaser, but yeah, basically, ⁓ choline that we consume is a very important micronutrient, is super important in the central nervous system and the myelin sheaths. And then ⁓ indole is actually one of these micro poops from E. coli that actually stimulates neurogeneration, not neurodegeneration, but neurogeneration or regeneration. ⁓ And there are many, many other aspects of microbial physiology in our gut that supports a healthy central nervous system. So I'm sure we will bring experts and discuss these topics later in future episodes. speaker-0: absolutely think we should do the connection of the gut microbiome to neural health, neurogeneration and the... Maybe we will, maybe we will. Speaking of Livy, I also wanted to show my Gen Z committee that inducted me and this is the gang. They're a bunch of cool people. speaker-1: Maybe we can have Livy as one of our guests. and you see- Looks legit. speaker-0: legit legit and look look at all of them doing the six seven thing with me so it's amazing I loved it I had a lot of fun and hey guys Gen Z folks I hope you can all enjoy the rest of this episode and I want to turn back to you Momo and say what was on your radar speaker-1: All right, so ⁓ I just had a super exciting trip to China. I went there because I was an invited speaker at a big conference, 45,000 attendees, and they invited me and paid for all my travel and really hosted a great, great conference. I gave a presentation and then I had a panel discussion on GLP-1 and all kinds of details about that. Besides the professional part, I really enjoyed visiting China. It's one of the most enjoyable countries to visit for many reasons. And whenever I travel, food is one of my major reasons for travel. And I really enjoyed Chinese cuisine. And it's interesting because I had not heard of Chinese crepes. But Chinese crepes are a huge deal. Like every street in China had Chinese crepes for breakfast. It's a huge deal. ⁓ tasted about six or seven different ones. And while I liked most of them, this particular one absolutely blew my mind. I absolutely fell in love with it. And so I researched it and using Chachi PT, I figured out exactly what it was and how to make it. And I fine tuned the recipe. And so I actually made that at home and I made it actually healthier than what they make in China. And so maybe in one of the cooking focused ⁓ episodes, we're going to actually show you how to make these crepes. But until then, Look up Tridys crepes and see if you can make some at home. Super exciting stuff. speaker-0: Awesome. This is very, very attractive looking and I'm gonna look this up and try it myself, Mama. Thank you for bringing that up. let's then go to the main body of our episode, which is where we're to talk about the wet lab and the dry lab of Vio. speaker-1: journey of poop and blood and saliva from customer samples all the way to data generation. speaker-0: Exactly. And in order for people to appreciate what we are talking about, I put together this little picture where we talk about the journey of the samples that people send to Viome and how that gets turned into insights and how it gets turned into supplements that can then be ingested to improve and potentially transform your gut microbiome in the right direction. this picture you see on the point number one, that's you. sending your biological samples, stool, blood or saliva. And that goes into our Viome CLIA lab, which is what we're going to talk about today. And Momo is going to explain so many things about the CLIA lab through pictures and videos and everything. It's going to be so cool to hear all of that stuff and see all of that stuff because it is like science fiction, many of those parts. And then what comes out of the CLIA lab is a digitized form of the samples, which are the physical molecules. that gets turned into digital bits and bytes and that goes into our bioinformatics, which we're to talk about how bioinformatics make sense out of all of the data that comes out of the lab. And then that information is scored through the Viome scoring engine. And then recommendations are generated through the Viome recommendation engine. Both of those are available in your app, which is right in the middle of this entire cycle. And the recommendations then go into a precision manufacturing facility that Viom has that generates precision supplements that goes back to you as the customer. And you have to take that for a length of time. know, typically people take it for up to six months and then they retest and the cycle begins again. So that is the overall process. But what we're to do today is break down what happens to your sample when you send it. So Momo, I'm going to... ask you the first question. So when I send you my samples, what does the lab do with it? speaker-1: Yeah, let's get started. So let's start showing slides and just give a little background. back in 2008 or nine, I started to realize that what was essential was to quantify the human body in multiple sites and to quantify it very deeply. But the point is that you have to quantify the right types of molecules. And so in 2010, I completely reset my scientific career, and then spent six years developing what we're about to talk about. And it's essential to understand that Viome is the only company that offers RNA sequencing tests direct to consumer. And it took six years to just develop the technology before we could even start Viome. And we're going to talk about some fundamentals here. So the first fundamental is let's talk about RNA and transcripts and gene expression. Let's tie that all together. So a typical microbial genome has about 3,000 genes. And each one of those genes has a function, but not all genes are active at the same time. And we're going to actually show you what percentage are active. But for now, let's just zoom into these three genes on the genome, knowing that there are 3,000 of them, but we're just going to look at these three. And what you can see is that DNA has one copy of each gene, but DNA doesn't do anything. As we discussed before, it's just a cookbook, but you don't get to eat out of that. You have to make actually the food out of that cookbook. When the bacteria need those functions, they're going to copy that DNA into RNA. So all of those green molecules and blue molecules, you see those are transcripts or RNA molecules. So transcript is just a scientific term for an RNA molecule. They're complete synonyms. And so you can see here that the green gene is copied a lot more times than the blue gene, and the purple gene is not copied at all. That means that whatever function the purple gene is encoding, that bacterium is not going to perform at all, and that the blue function is going to be performed but at a low level, and then the green function, the bacteria really wants to perform a lot of. And so this is really a key to understand that the gene expression level determines how much of that micro poop a bacterium is going to produce, which is then going to affect our physiology. So I want people to tie the level of RNA, the number of RNA molecules to the effects on human health. everything that we're going to talk about in the next 30 minutes or so talks about how do we quantify these RNA molecules very accurately, ascribe them function, quantify them so that we can see how much my microbiome is contributing to my health versus yours in thousands of different functions. speaker-0: Yeah, Momo, before we go to the next step, I wanted to just also ask you about how these RNA copies are actually performing their work. Is it directly on the end molecule or do they get transformed again into proteins in some cases and no protein in some other cases? speaker-1: Exactly. there are, so RNA is such a versatile molecule. And when we get Tom check in, we're going to talk about all kinds of functions. for now, I would say, ⁓ just think of that RNAs perform two general functions. One is non-coding, which is regulatory functions. So regulatory RNAs, they will regulate other physiological processes. But What most people think, and this commonly happens, is that the RNA is actually then translated into a protein. And the protein is the one that actually performs the chemical function that produces micro poop. So when we talk about butyrate as being micro poop from bacteria, there was a series of proteins that converted fiber into butyrate. And those proteins were translated from multiple transcripts. it. speaker-0: Okay, very good. So now when you talk about a lot of copies of one and a few copies of the other, I want to understand, are all microbes active at all times or is only a few microbes that are active? And if the ones that are active, they only express a few genes, right? So what is the difference between the potential and the actual? speaker-1: Yeah. So we have a couple of slides that we're going to show now, but yeah, I would like people to think about our poop containing on average, just over a thousand different microbes. And then remember that each microbe encodes on average 3000 genes. So if you multiply those two, we're in the four to five million genes range, meaning that a typical gut microbiome and codes four to five million functions. is incredible because the human genome encodes depending on how you counted tens of thousands of functions, but definitely not more than just, say over a hundred thousand functions are encoded by the human genome. Whereas these microbes are encoding millions. So of those microbes, there are some that are very active depending on what you feed them. And there are some that are much less active. And so if you look at each microbe has certain level of activity and we quantify that. And then within each microbe, They will only express the genes that are needed for them to live and thrive in the gut because there's a energetic penalty. Whenever you're transcribing a gene to make RNA, it actually costs the microbe energy. This doesn't come free. It's a very energetic process. And so the microbes want to make as little RNA as possible because if they make RNA that they don't need, they're now less competitive in a highly competitive community. And so they're only making the RNAs that they need. So let's now go to the next slide where I think we're going to explain. Yes. So here we're showing ⁓ in the blue graph, how many genes are encoded by a typical gut microbiome. And the red bar, the red bar is showing how many are expressed at one part per million or more. So one part per million is an extraordinarily small contribution to the community. And yet, even if we set the bar that low, you can see how few genes are actually expressed. It's on the order of a hundred thousand, let's say. And so this is where, ⁓ when I commonly say, you know, 50 % of human biology is microbial and 50 is human. I'm thinking that at any given time, about 50,000 functions are expressed in the human and 50,000 functions are expressed in the microbes. that's about a hundred thousand, but we're definitely going to have an episode diving deeper into that. But this just shows you that. those other microbiome tests that are doing, let's say, metagenomics, these are fancy metagenomic tests, all they can tell you is your microbiome encodes four million genes. And based on that, you may be tricked to think, wow, my microbiome is super rich. It's performing all these functions. Well, guess what? If you don't feed your microbiome fiber, it's not going to be producing butyrate. It cannot produce butyrate. It doesn't matter that... that it encodes genes that are capable of producing butyrate. Unless you feed it and unless it's actually expressing the genes for butyrate production, you will not get any benefits of butyrate. And so this is really the power of Ion tests that I cannot overemphasize that we're not only measuring which genes are expressed, but their level. And we're going to dive deeper into that in our bioinformatics. speaker-0: Yeah, and the other observation from this chart here is that the total percentage of expressed genes is less than 5%. You see on the right-hand side, the mean or the median is going to be approximately 2 % and even the high end of it is like 3.5%. So it's less than 5 % of the total amount of microbial genes are expressed at any given point. Yeah. speaker-1: Yeah, I mean, that's a good point. I just feel bad when I read a paper where they use metagenomic sequencing, and then they use all genes that they found for machine learning, and they find all these machine learned features in one paper. And then in another paper, they find completely conflicting information. And the reason is because 98%, as you mentioned, 2 % of the genes are expressed. 98 % of the genes are silent. but they're counting them in their models. So 98 % noise and 2 % signal. That's not good for machine learning. As you know, machines will find something. So let's go now to saliva. I mean, it's a similar story. You can see on the left side, how many genes are encoded by a typical salivary microbiome. And the red bar shows how many are actually expressed at any appreciable level. And then the percentage is shown in the graph on the right. So similar story. speaker-0: story. Okay, yeah, I want to get back to the lab and I want to understand what happens in the lab. So tell us please, how, when I send my samples in, what is the first thing that happens in the lab? speaker-1: Yeah, let's go over the whole process. I think my next slide shows sort of high. Yeah, this is an overview of the whole process. And then we'll dive deeper into some of these sections because we have some videos and slides. So I'm going to go pretty quickly over this and just highlight one thing here. So we, we receive these stool blood and saliva samples from our customers. The first thing we do is we break up all microorganisms in them. That's called a cell lysis. And we use mechanical and chemical lysis to break up gray and positive cells. fungal cells, human cells, every single thing that's in there, we break it open and we go. speaker-0: And we're gonna show a video of that. speaker-1: We're going to show a video of that, then now we are going to purify nucleic acids, is DNA and RNA. And then we're going to degrade DNA because as we just saw, DNA is not very informative. And so we're going to get total RNA. Unfortunately, total RNA of any sample contains mostly non-informative RNAs. These are ribosomal RNAs, which... Unfortunately, 16S tests for microbiome sequence, but they're so non-informative that we actually remove those. And so we physically remove ribosomal RNAs and a few other transcripts. Now our audience knows what a transcript is. We remove a few of those that are super highly abundant, but not very informative. And this makes our process such that we can actually translate this test into a consumer product and people can afford it. You know, if you're doing a scientific study, you know, these Metatranscriptomic analyses can cost $1,000 per sample, whereas here for us, we can bundle three different metatranscriptomic analyses and charge customers $400 or less and still do fine. And that's an extraordinary achievement. And this is one of the reasons why. And now we get into a really, really highly technical aspect, but I still want to discuss it a little bit. And maybe we'll cover this in more detail in another episode. now that we have the actual useful or informative RNAs, we want to sequence them. And we are going to convert them into something called a sequencing library. And sequencing library must be made in order to be sequenced on a sequencer. And this library is made using a custom process at Vallejo. But one part that I want to highlight is that our laboratory is the only one that uses this special patented barcode hopping method where we actually have a 10,000 fold higher fidelity than other methods of sequencing. And this is an extraordinarily important bit because when I used to do sequencing at Los Alamos National Lab before this invention, we used to get 1 % noise in all of our data and it was very difficult to interpret what was actually noise and what was the signal. So we worked very hard for several years and developed this method of creating a library that moves that noise from 1 % to one part per million. So we dropped it by 10,000 fold. And now when we measure things, but we measure, for example, quantified transcripts above one part per million, we're purely measuring a signal. And this is a quantum leap over every other test on the market. And we don't really highlight it enough because it's kind of really highly technical. But it's super important that we get that. speaker-0: Yeah. And Momo, as a machine learning guy, I got to say that if you had 1 % noise, we'd be picking up many, many features below that noise floor. And we'd be picking up essentially a useless or meaningless kind of a model. But now that you've dropped that noise floor down to one part per million, we're going to pick up a number of features that are above that noise floor. And therefore it's going to be more meaningful. speaker-1: Yeah, that's really the key. And I'm just super proud of that achievement. that's really important. speaker-0: This whole process of library prep and all of this stuff happens in an automated fashion in our lab, correct? Correct. We're going to show a video of today. speaker-1: Yeah, this is fully automated in a CLIA lab and we'll go over the details soon. And yeah, the last step is sequencing. So once we prepared the sequencing library, we do quite a bit of QC and then we load that on the sequencer and we're using a state of the art sequencer that we're going to cover in a few minutes. So that's the process basically from sample to generating sequencing data. And then we'll have a special guest to break down for us to bioinformatics. So let's go to the next slide. speaker-0: some. speaker-1: And obviously the audience is welcome to ask any questions in the comments. We welcome any questions and we'd love to address them. Okay. So here are the samples when they come in. These are obviously we get thousands and thousands of samples, stool, blood and saliva. And this is what they look like when they arrive. So let's go to the next slide. speaker-0: is the stool, the green is the saliva, the pink is the blood samples. speaker-1: Okay, yeah, let's talk about this a little bit more because there is a there is a process here. So yes, the brown ones are stool and green ones are saliva and those are not automation ready. The pink rack has these white tubes and they're actually automation ready as they are collected by the customers. So those come into our lab completely automation ready. So let's talk a little bit about Clia right now while we're here. So we run a Clia laboratory, that's a clinical laboratory standard, both created and overseen by the US government. And so as a CLIA laboratory, we have a quality management system that basically prescribes every single thing that happens in the lab. And it's overseen by the state of Washington, the state of California, the state of New York. And those states oversee it and make sure that we're doing exactly what we're supposed to be doing to generate high quality data. So now, the first part of the process is what we call sample receiving. Every sample has a barcode, and we scan that barcode, and the system recognizes it as, OK, this sample was registered by a customer. And then the accessioning team looks at the sample and either passes it or rejects it. If it passes it, it'll go to the laboratory. If it rejects it, we actually ship a replacement kit right from the same lab the same day so that they can recollect the sample if they made a boo-boo collecting it. So those samples that pass QC, Now go into our aliquoting lab where stool and saliva samples are aliquoted or split into smaller tubes, just like the ones you see on the bottom for blood. And these are going to be now automation ready. basically all samples end up in the exact same tube. And then from there on, they're all treated the same way. it. speaker-0: And what about this process of transferring, you know, all the quality control and the transferring of aliquots, all of that stuff is managed through a laboratory information management system so that we can keep track of all the different steps and the different functions, the specific samples, the aliquots, everything else, correct? speaker-1: Yeah. So we have a state of the art limbs software. So laboratory information management system software called labware. And it absolutely does. It manages all of our lab operations in terms of keeping track of everything. So everything is barcoded. Every reagent is barcoded. Every sample is barcoded. Every process is barcoded and everything is recorded. And that's a part of a clear process to make sure that your sample tracking is such that you cannot mix two samples. and then at the end provide mixed results to different people. speaker-0: Exactly. The standard operating procedures, SOPs, are documented or trained for each of our laboratory technicians, and then they are inspected and validated and checked off every so often, correct? speaker-1: Yeah, exactly. SOPs are developed in the R &D stage and then they're validated clinically. And once they're validated clinically, our staff is trained and we perform it every day based on approved protocols. The lab director has to approve every protocol before it's implemented into CLIA lab. speaker-0: Yeah. All right. So what are we going to see next? speaker-1: Okay, now we're going to see this bead beading process. So this is the cell lysis process. And this machine is called a bead beader or a bead basher. And it's basically a very fancy instrument where we now have our samples inside these little automation tubes. And inside those tubes are these beads. They're microspheres that are a few microns in diameter. They're very high density and they're very tough. And basically by shaking the tubes at very high speed in all axes, x, y, and z axes. You're basically moving these beads about 20 miles an hour through the sample. And when a bead runs into a cell, whether it's a bacterial cell or a fungal cell or a human cell, it physically blows it up. And the chemical treatment of the sample already weakened the cells. And so this process is very important in that. we're actually lysing or destroying all cells in an unbiased fashion. Whereas many other tests ignore this process as sort of, let's just do some cell lysis and call it whatever it is. This is a clinically validated process and it's very uniform and very nonbiased, which was a very important part of the whole process. speaker-0: Amazing. So after the lysing step, what is the next big one? speaker-1: Okay, so after the lysing step, we also centrifuge the sample to get rid of all the cell debris. So that goes to the bottom of the tube and we take the solution from the top called clarified lysate. And now it's going to go into a machine called Kingfisher and Kingfisher extracts fully automatically total nucleic acid. So here's Kingfisher. It's got a turntable with a bunch of plates and reagents and it's got a magnetic head. And you can see now we're already in the 96-well format. So every single thing from now on until sequencing is done ⁓ in a 96-well format. So whatever we do to one sample, we're doing to 96 samples. And you can see now the Kingfisher is actually mixing these special silica beads that bind RNA and DNA in the clarified lysate so that we can purify them. And so this process lasts about 40 minutes, and we extract basically DNA and RNA, and we remove proteins and lipids and polysaccharides and all other molecules from the sample. So we purify nucleic acids. speaker-0: So Momo, how do you know the quality and the output of each of these steps is going to meet the requirements that you had set up? speaker-1: Yeah, there are intermediate quality control steps that are all validated. so everything is clinically validated to ensure that we get high quality data at the end. And we're going to touch upon the details towards the end of the podcast with our bioinformatician. We're going to talk about quality control, but right now as a sort of intro, I'd like everyone to know that every one of those plates with 96 samples contains positive process controls, negative process controls. and correlation controls, as well as every single sample gets an internal control as well. And so we can do batch level quality control to make sure that the instruments and the reagents worked. And then we can do sample level quality control. And we'll go into some more details towards the end of the podcast. speaker-0: Great, so now we have created this 96 sample batch and we're ready for the full sample prep process. speaker-1: The library preparation. Now we're going to go to the library preparation process. this is where a very, very large, very expensive and super precise machine called liquid handler handles the rest of the process. And so these are Swiss made, absolutely highest quality liquid handlers. They can handle volumes down to a couple of microliters. And this is what we call a pod. A pod consists of speaker-0: Yes. speaker-1: two robots or liquid handlers, some accessory equipment, and one laboratory technician manages these two robots. The laboratory technician feeds the robots, all the plates and all the reagents, but the robot does all the work. So everything is automated there. The technician doesn't actually do any pipetting or any liquid transfer. So let's now go. I have a video that I took of the robot in action, just so that you see, the audience can actually see what this robot does. So this is what we call an independent channel head. So this robot has a 96 well head called a core head, and it has an independent channel head. are eight different channels that can do their own thing. And so there are some places where depending on the well, each head will, each channel will do its own thing. And so you can see here how let's play it again. The robot picks up eight of these tips, plastic disposable pipette tips. It will go to a reagent plate. and pick up that reagent. It will then move it to the sample plate where our samples are. It will dispense and mix. It's now mixing that reagent with the sample. And then after it's done mixing, it's going to dispose of the tips. And this is one of the rare steps where independent channel head works. It's actually the core head that's doing the vast majority of the steps. So this is where we're degrading DNA. And this is where we're doing that library preparation. where we're converting RNA molecules into DNA molecules that are going to end up on the sequencer, and we're providing them with molecular barcodes that enable this high fidelity of 10,000-fold improved high fidelity where we're making sequencing libraries with such high fidelity. speaker-0: So Momo, you explained it in such a simple way, but my head spins when I think about all the stuff that's going on, because I remember the different steps that you explained earlier that are all compressed into that one liquid handler, right? So if I remember correctly, what you had was the removal of the non-informative RNA and then amplification of the informative RNA molecules. And then you're doing the barcoding. and everything else that's required to put the 96 samples in a form that can then be taken to the sequencer. speaker-1: Yeah, exactly. So the PCR step is an amplification step, but normally it's used to amplify the sample because you don't have enough sample. We actually have enough sample. So we use PCR for a different purpose. It serves two critical purposes. One is during the process of PCR, we are actually incorporating these molecular barcodes into each sample so that after PCR, we can mix all samples from thousands of customers and sequence them all at once. And this is what gives us a huge cost advantage. And the second purpose of PCR is we actually have a patented PCR process, which I'm very proud of, which actually evens out all samples. So you see, when you load 2,000 samples on a sequencer, the sequencer doesn't know that they came from 2,000 people. It doesn't know anything about barcodes. It simply sequences all molecules. But if you loaded 10 times more sample from one customer than another customer, that first customer is gonna get 10 times more data and we don't want that. We don't want uneven data coverage. We want even data coverage. And so in order to do that, we use this very special PCR that evens out the game, meaning those samples that already have enough material in them, PCR doesn't basically do much. Whereas those samples that have less material, it brings them up to the same level so that all samples have the exact same amount of material at the end of PCR. And now we just pull them all together. And then we can load them onto the sequencer and sequence them all in parallel. Got it. speaker-0: got it. That's just mind-blowing for me that all of that stuff is happening in that one liquid handler. speaker-1: It's absolutely amazing. And by the way, this is patented. So, and while the previous patents we discussed in terms of molecular barcoding and depletion, those were all developed at Los Alamos National Laboratory and we exclusively licensed them. This particular patent where we're using PCR to even out, you know, a hundred customer samples or thousands of customer samples, that's actually solely a Viome patent. We developed that at Viome and I'm very proud of that as well. Amazing. speaker-0: All right, so the next step is the magical step, right? It's the ⁓ sequencer, correct? speaker-1: It's one magical part of it. All of these steps before are still magical because there is no entity in the world, no government, no company, no university, no institute that has this process worked out. No other entity. And it's amazing that we patented the core technology 10 years ago, more than 10 years ago, and still to date. There is no other entity that has figured it out. That's how complex it is. ⁓ And a part of that complexity is yes, this next step, is absolutely mind blowing still. I just, yeah, I'm just, it's like as if you went on a spaceship and you're going the speed of light, like how you can get used to that. That's how impressive it is every day. So yeah, so now. that we have generated sequencing libraries, we actually need to read those molecules and generate the sequence. And that's done by this NovaSeq X plus instrument from Illumina. And we have to thank many people and many technologies for where we are today at Vion, including all the computational developments that you're gonna discuss in the Boninformatics pipeline. But we have to be grateful for so many things. And this instrument is one of those major keys to our success. This instrument... will read at the same time in parallel 10 billion molecules in sort of a medium scale level, but it can read 25 billion molecules in a larger scale. And then because it has two different separate parts, the maximum throughput of this instrument is that it can read 50 billion different molecules all at the same time. mean, literally that's not an exaggeration. Wow. At the exact same time, it'll read 50 billion molecules. Yeah. Let's talk about that a little bit because, I mean, this is unbelievable. So let's go to the next slide. Yeah, so this is the core of, so that instrument is a huge box, but everything else is sort of supportive. This is where the rubber meets the road. This is called the flow cell. And if you can see, it has eight little vertical strips called lanes. So it has eight lanes. And so this instrument can run two flow cells in parallel and each flow cell can sequence up to 25 billion molecules. We're currently using a flow cell that sequences 10 billion molecules in parallel, because that's our current laboratory throughput. And so I'm going to dive a little deeper here, but I would encourage the audience to go to YouTube and search for a video called Illumina Sequencing by Synthesis and go watch a few videos. It'll blow your mind because... speaker-0: Let's put a link to that in our show notes. speaker-1: Oh, that's a good point. Yes. I will put a link to that YouTube video in our show notes. Great idea. So in 2004, we were still sequencing DNA and RNA using Sanger sequencing. And this was the technology that was used to sequence the first human genomes. And at that time it was state of the art. But the instruments that were used at that time, their maximum throughput was 96 reads at the same time, meaning they could read the sequence. of 96 molecules at the same time. That was the max. And this instrument reads 50 billion molecules at the same time. I mean, that's a billion fold improvement. And so in 2004, if someone told us, hey, in 15, 20 years, you're going to be able to increase this throughput a billion times. There's nobody on the planet that would have said, yeah, that's going to happen in my lifetime. That's just like saying, that in 15 years you can sit in your car and drive the speed of light. Like, who's going to believe that? That's not possible. But Illumina made it happen. And that's, like I said, one of the key reasons why we have been able to put together what we have is that we can sequence to such a large degree at such reasonable cost. speaker-0: Yeah, yeah, whenever you say that, I also think about Moore's law on the computational side, which has basically doubled the capacity of microprocessors every 18 months or so, right? And that has been going on now for more than 50 years. And so I remember when I was going to college and if people would tell me that the size of mainframe computer back in those days could then be compressed into a small device that you could put in your pocket and Today, turns out an iPhone has more compute capacity than the big mainframes we had back in the 80s. Right? So this is a similar thing. Maybe it's even more than that because it is taking the physical world and it's turning that into the digital world at scale. You're talking about 25 billion molecules and each molecule is many, many nucleotides. Correct? So we are talking about a humongous amount of data that's being generated by a single flow cell. speaker-1: Yeah. No, I really think we need to highlight both experimental and computational side that both of those have enabled us to do what we do today without either, there would be no point in generating this much data if you don't have the computational power to analyze. So yeah, unbelievable progress in technology and just super grateful that we live in these times. speaker-0: Tell us a little more about what's underneath the covers of these flow cells, please. speaker-1: Yeah. So let's go to the next slide so we can zoom, really zoom in. So there's on the left side, that flow cell that has eight lanes and each one of these lanes has about 2 billion nano wells. And I've shown that structure, what it looks like on the surface. And each one of these wells is going to actually produce one sequencing read, or in our case, two sequencing reads from one molecule because we're doing paired end sequencing. And we're going to talk about that a little bit in the bioinformatics. part later on. But yeah, Illumina has been able to fit 2 billion and more, in the case of the larger flow cell, 5 billion of these nanowells in one lane in order to do that parallel sequencing of that many molecules. speaker-0: Momo, this is like nano lithography, right? So there are machines probably that are at the nano resolution able to dig those wells and make sure that they have the right shape and the right size and everything else so that this whole process can happen downstream, correct? speaker-1: Absolutely. It's nanoscale. It's unbelievable, basically. Like I said, in 2004, this would have been science fiction. You would simply say that we can't do that. And yet, it's happening every day. And it's now a routine. It's not like we have to do anything special. The instrument is very easy to operate. speaker-0: It's also parallel to the silicon world. just want to mention that, right? the Moore's law thing that I was talking about, there's also nanolithography that happens in these fabs that result in all the current generation GPUs and CPUs and everything else that happens at the nanoscale. are many structures like transistors and capacitors and resistors. All of those things are being built on these chips that do similar types of very, very high resolution lithography. And I think it's a similar type of scale and process by which these technologies are created. speaker-1: Yeah, lots more to unzip here, but that's all we're going to cover in this podcast about the sample to result in the physical world. And if audience has questions, like I said, they can either post comments in our episode or they can email us at podcast.vion.com and we'd be happy to explain in any more detail any of these pieces. speaker-0: Yeah, so at this stage, what we have done is we have put all of those libraries that we prepared in your liquid handlers and we put them into these nano wells in the flow cell and we literally put that into the sequencer and the sequencer does what again to convert that into data? speaker-1: Yeah, so the sequencer will read, and this is that video that we're going to put up, is sequencing by synthesis. It's going to physically read the sequence of each RNA molecule or each transcript and provide that sequence to us. And so we will link that video. speaker-0: Excellent. And that brings us to the transition from the wet lab to the dry lab, right? The dry lab is where the data that comes out of the sequencer is transformed into meaningful, I would call it insights, right? So starting with norms of the genes and the functions of these microbes and human organism, and then all the way downstream of that. I think we're ready to switch over to that dry lap segment now. speaker-1: Okay, so now let's talk about the very last step in the physical world. The last thing that happens in the physical world is that the sequencer is translating the sequence of the physical molecules into digital data with the proper sequence of the bases and uploads it into the cloud. And it does that billions and billions of times, and it generates these very large files. And now from here on out, everything is in the virtual world up in the cloud. where we're going to dive into bioinformatics. speaker-0: Awesome. And in order to talk about bioinformatics, we have our own Dr. Who, Dr. Lan Who. Welcome to the show. You look really nice. What is that hat or where is it from? speaker-2: Well, Doctor Who, every season there's a new Doctor. This is my favorite. The fourth Doctor Who by Tom Baker. And then I found this Fedora hat. And also, right after I finished my PhD, I knit this Doctor Who scarf as my trademark. speaker-1: Amazing. Absolutely amazing. speaker-2: Greetings, yes! speaker-0: You're a fool. So Lan, you are our resident bioinformatics expert and you live the world of all of this data that comes up from the sequencer into the cloud. And we have this little picture of how the data looks like when it is uploaded into the cloud. To me, it looks like gobbledygook. It's just like a lot of A's, C's, T's and G's. and it's organized in some fashion that I don't even understand. has all these I's in the middle and pluses and all of that stuff and there's a humongous amount of this. How do you make sense out of this stuff? speaker-2: Well, so what you just gone through, this a way that the sequencer writes out about a read. A read, as the screen shows, it's 150 letters long. That's determined by the chemistry to make the library that Momo already explained. And then you kind of see this patterns in this little screenshot. So every four line describes a read. So this is from the physical world translating into this digital world representation. So the first line, it tells you about the flow cell, which particular lane, which particular barcode. And the second line is ATGCs. This is actually the sequences of that particular fragment of the RNA molecule. And the third line, it's a giveaway. It used to be a repeat of the first line, but then it just doesn't really add information. The fourth line, a lot of I, I, I, I's, that's a ASCII code, which is a letter encoded representation about the quality of that particular base. When it turns from the fluorescent into the image and then turn into a base. So that's the score for the quality. So you will see every four lines, every four lines, every four lines. And our sequencer speaker-0: Got it. speaker-2: produces two of these files for every sample. So that's the paired and sequencing. Paired meaning two, know, one and a two. So for each stool or saliva samples, there's these two gigantic text files, about 60 million lines per file. So taking all together, every sample would produce somewhere between eight to 10 million these short reads. represented in these two files. speaker-0: Yeah, okay, let me ask you one more question here. You're talking about these things called reads. Why are they called reads? And what are the properties of a read? speaker-2: A read, it's the, you could just think as when you walk along this long string and you read every chunk, this one chunk is 150 bases. So that's why it's called a read. So just imagine as a readout of a much longer strings of these letters, which underline that's the RNA molecule. speaker-1: Yeah, the read is a shortcut for sequencing read. And it's basically what the sequencer reads in terms of the sequence from the physical molecule. speaker-0: Okay. So can I think of it as like, you know, a long, long, long string that's been chopped up into little tiny strings and each tiny string, which is approximately the same length is essentially a read. And now it's all jumbled up in this big file of 16 million lines per file. And there's two large files per sample. So it's a huge amount of jumbled up little strings. Now we have to make sense out of the whole thing. So, Lon, how does bioinformatics help us make sense of this jumbled up collection of little strings. speaker-2: Sure, so this is like, you you have the book, you chop it up and throw this into the air and now we pick up a little small fragments and try to stitch these little fragments to reconstruct your original story. So if you go to the next slide. ⁓ So first, before we make sense of it, we need to know what we are looking for. And for our purpose, we are not doing discovery. We are doing the detection. So by detection, we need to have a menu first listing all the possible things that could be detected in a given sample. So what is here showing is the very simple schematic representation of the menu. We have, ⁓ let's say, each Little bug has its own genome, which is a very long stretch of a DNA sequences, ATGCs and representing this one gray bar. And we have many, many, many of these bugs. And each colored box represents a gene, which is the block of these DNA that codes for something to carry out a function. The color coded here, just to say, even though they are coming from different genomes, But they need to carry out very essential functions, which are actually very similar across the different ⁓ microbial genomes. So there's different color blocks representing the same type of functions. Now, there's another important aspect ⁓ if you click this slide. So that's how we describe it, as I just went through, ⁓ different Genomes could code for the genes that actually do the same thing. So you will see the red function, blue function, and then the green function, for example. The other is that we constructed this tree of life because we, ⁓ different genomes, we would like to know how they are coming from either the bacteria or IKEA or fungal, these larger branches on this tree of life. So we first taking publicly available resources, NCBI National Center of Biotechnology Information. That's our go-to source and it's very well respected and all that. And we take the highest category in terms of quality in that collection, which called RefSeq, short for reference sequencing. So as the name entails, it's for the reference. So that's ground truth. So we take the collection of genomes from that category, and we also take the complete genomes. So this whole repository, reference the sequencing repository, has many levels. The complete genome, which is the highest, and then depending on the prior work, they may not have the complete genome. They may only have the chromosome level, so on and so forth. But we... only take the complete genome from the referencing sequencing this particular high quality database to construct this tree of life. So you might hear numbers, you know, crazy tens of hundreds of thousands, all these genomes. it really devils in the details. So the devil here is that we are taking the most complete collection from the most respected the reference sequencing database. of NCBI to construct this tree of life. speaker-0: I have to ask again, the reason we do that is because we want to ensure that the output data that we provide to our customers are of the highest possible quality. So we have to start with the highest quality data, both on the catalog side, as well as on the sequencing side, and we match the two and that's how we get the highest quality data, correct? speaker-2: Yes, so what I just explained is this Guru said it's catalog. So it's our menu of things that we will be detecting. Now, once a particular sample is down sequencing with those tens of millions of these lines from those two gigantic text file, we will have an algorithm running to grab those each of these little short 150 letters long read to map. into this collection of genomes and the genes and the functions. So the algorithm, I don't need to go into the details, basically it does it very efficiently. At the end of that, for each of these manual items, we get a read count. So if you go to the next slide, that's how we're showing each little bar in the bottom. The short bar represents that 150 letter long read. So the algorithm would do, it will map whenever it can into this whole catalog of items. So in this example, genome 1 has all these bunch of reads mapped to little black bars, and then each gene will have a certain number of reads mapped to. And then basically at the end, we do the tally. So we will have a number of reads mapped to the genome 1, which is a collection of all these little bars. And then Genome 2, the same thing. Yeah, exactly. So the read counts to the microbial gene, which is that little green box. And then the second, the longer dashed line box, that's the recounts to the microbial genome. So it's a collection of all the reads mapped to that particular genome. And then the next for the function, now this is the key part. So the genome gene, it's easy to understand. A lot of microbiome. algorithm would do that or other detection technology would do that. But we are detecting RNAs. So RNA doing each RNA molecule would do something. So here showing all these green bars, even though they are coming from different bugs, but they are doing the same biochemical function. So we combine all these together to represent that particular microbial function. So think of as this It's everybody contributing to this pool. Once this thing is in the pool, it doesn't really matter who provides it. So this class of microbial functions, it will just be carrying out a certain particular biochemical reaction in the body. speaker-0: Great, great point. So Lan, tying this back to some of the stuff that we've talked about in the past. So if we are talking, for example, about short chain fatty acids, right? So let's say it's the generation or production of butyrate is one of the examples we used in a previous episode. So the green could be butyrate production, but the butyrate production function is happening in multiple different organisms, which are these different genomes. And what we end up doing is we don't count the genomes, rather, we count the number of expressed functions from all of the genomes and we combine that together and say, this is the total amount of butyrate that's being produced in your gut across all the organisms. That's a fundamental concept that is quite different in what we do compared to all the other or most of the other microbiome research that you may be reading about. speaker-2: Correct. So we do this vertical in this little chart. It's the vertical combination to represent a microbial function. A lot of people do this horizontal way, which we do as well. So we got both. speaker-0: Yes, yes. speaker-1: I want to just add a couple of numbers here. think people would be interested to know that an average bacterial genome is 3 million letters long. So 3 million bases long nucleotides. And that an average bacterial genome has about 3,000 genes. And so this is just a tiny little sliver of a genome. It's far more complex than this. speaker-0: Well, speaking of numbers, Momo, I want to actually talk about the actual number of genomes we detect in genes, we detect in functions, we detect in the sample types. And I'm going to get there, but before I ask you that question, Lan, I want to ask you one more thing about you mentioned algorithm, right? These algorithms are pretty sophisticated algorithms. They have a ton of work they need to do. In fact, it's one of the most advanced, you know, sort of the state of the art. type of algorithms that takes up a huge amount of compute and memory resources on the cloud, correct? So do you want to say anything about the, I would call the power and the sophistication of these algorithms? speaker-2: Right. So these algorithms have gone through quite some evolution since this field is established. For microbiome now, compared to human genome, the microbiome each genome is short. But we wanted to look at tens of thousands of these things at the same time. So it's a very different type of problem. Human genome, it's a lot. It's 3 billion instead of 3 million. But that's only one. So we are talking about tens of thousands. So to parallel this whole process, that's the key. So we, once we, the whole flow style, you know, all these gigantic files are up running in the cloud. We put that into multiple chunks we call. So each chunk could kick off its own pipeline. So we parallelized as much as possible that in a way to shorten the whole process time. So the Another thing is, so while we want flowcell, there's thousands of samples going in. Each sample generates that two big files. It's just the amount of data coming out to one sequence and run is really mind boggling. And then even with all these parallelization and optimization, it still takes about 10 to 12 hours to process one flowcell. speaker-0: Wow. speaker-2: And that's already with a lot of the smart brains behind it how we make it faster and then to paralyze as much as possible without crashing a physical machine limit. Yeah. Computer machine limit. speaker-0: Exactly. And speaking of physical machine limits, know, there have been situations where some of the leading clouds that we run on have basically reached the limit of their own memory size and the compute size and the storage size and so forth. So it's very important for our audience to understand that these are extremely large, complicated computational workloads that require state of the art. computers, clouds, and in fact, it costs quite a bit of money to run these workloads on our cloud, right? So given that, ⁓ Lan, I would like you to also take us through some of the numbers. So here are the numbers. Please tell us how did we end up with these numbers? speaker-2: ⁓ so again, you know, this is on average across many, individuals. So don't take it too literally. I'm just giving you the sense of the amount of data, right? So from the stool samples, the on average microbial genome, which is the those the bars in the previous little chart, it's about 1900. And then the microbial genes, that's a lot more. It's over 217,000. Even though, so you were thinking ⁓ bacteria, the microbes, they're very efficient. They pack these genes very tight together because all their job is just to spread. But they still need so many ⁓ essential functions, building blocks. But now you're looking at the microbial functions, the number drops from genes because there are many genes from different genomes that are carrying out the same function. So you can see really the In terms of biochemistry, these essential, for example, the enzymes, it's just shared across a lot of microbial organisms. think we need to emphasize this really as much as possible because these are the things that are doing the work, not just something to hang out. These are actually the workers. speaker-1: We can also hang out. speaker-2: They hang up. But these are working building blocks, not just idle ones. So the saliva, it's very similar. then the other, well, the similar in terms of the microbial functions, but the genomes and the genes are less because, you know, with the GI tract is much longer than our oral cavity, so to speak. And then obviously just collect more and a different species will grow. But the interesting part is when you look at the microbial functions, they still need this many different things in order to survive and to replicate either help you or harm you. So it really depends on what the environment these microbes are living in. speaker-0: Yeah. And that makes the point that, you know, it's more important to look at the functions than the organisms. And in order to look at those 3000 plus functions, you need to first identify all the genes and then cluster them in the right way and annotate them in the right way, and then come up with the functions. So before we end this segment, Lan, I wanted to ask you a couple more sort of higher level questions. So, you know, we talk about the quality of the... samples, the quality of the sequencing, the quality of the bioinformatics and everything else. So first of all, how do we ensure the quality of the data that we get from the sequencing samples? speaker-2: So there are many levels. So coming out of the sequencing, as I explained in this gigantic text file, each base has a quality score. So we will look at overall how it's performing along the read. So if the enzyme getting tired or used up, the quality will drop. So that's the first thing. You know, we have a threshold, everything at least needs to be a moment correctly from around 90 or up. So there's this per base quality scale. Now, taking together, next level is the per sample. So our lab is a clear certified for each sample type. There's a set of ⁓ quality management metrics to check against. So every sample has to match all those metrics. And then they are tracked in the periodic review and all that. So this is a sample level. Now, we put all these 90 six samples in one batch. So that's the operational unit. So from batch to batch, each sample type on the batch, it has a bunch of controls. So there's a negative control, there's positive control, and the other one is the correlation control. So the correlation control is basically a real specimen, but it's repeated on all these batches. So from that correlation control, we will compare to its own representative reference, which was defined when this lot of correlation control, this batch of correlation control is released for production. the point being for the correlation control for each sample type, there's a reference. And from every batch, that particular correlation control will have to be very close to that reference. If it's outside the reference, the whole batch will fail. So we have these multiple levels of controls. Control to put in place. speaker-0: Wonderful. That's excellent. So you mentioned CLIA, right? So unless we meet the criteria that has been set up by CLIA, we will not be certified as a CLIA lab, correct? speaker-2: So the whole processes protocol is being reviewed, being inspected periodically. And in the operation, we have to match that. speaker-0: Yeah. Yeah. So it's, know, we say what we're to do and then we do what we're going to say. speaker-2: Do it and we show we actually did it. speaker-0: We show that we actually, so when some inspector from CLIA comes in, they can see that we have trained our people in the right way, we've set up our instruments the right way, we're following the standard operating procedures, SOPs in the right way, and therefore we are checked off, signed off as you're meeting the quality standards that are required. And that's very important for all the people who are using the Viome service, that we maintain that quality and that it's in fact inspected and signed off by a third party. in this case, CLIA, that says that we are meeting all the standards that are required. So it's a very, very important point. So ⁓ I think those were the main questions I had. Momo, did you have any further questions for Lan before we let her go? speaker-1: We can geek out all day. I want to actually just add a bit of a highlight that I am particularly proud of. So I think that medically relevant fungi are sort of the next frontier in our understanding our interactions with the microbial world. And I think that particularly molds, but even some other fungi play a significant role in chronic diseases and cancers. And this is not just a hunch. There's actually quite a few publications. However, Fungal genomes are notoriously dirty. They're notoriously fragmented and they're notoriously not reliable. So what I want to highlight is that we spent a year at Viome with Bioinformatics team combing through these genomes and cleaning them up and incorporating them into our catalog. So now we have about six or 700 high quality manually curated fungal genomes so that we can actually start to understand the role of fungi in our health. For example, one result that we have not yet published is that when we built a machine-learned model of brain fog, which is super common these days, the top features that came out of that model were related to fungi, not bacteria. And so we really need to, we have an opportunity at Biome to understand the role of fungi, and this is one of many, many ways that we can do that. So I'm very proud of that effort. speaker-0: Exactly, overall, the focus on quality is very high in our entire process. So you're talking about the quality of the genomes, which we use to map the reads coming out of the sequencer, but the reads themselves are high quality. And then before that, the sample preparation process is high quality, as we described earlier, the bioinformatic controls and all of the checks and balances that we have at different levels like Lon explained, they're also high quality. So all of these things put together contribute to the high quality of the results that we can stand behind. So wanted to make sure that that point is emphasized. And I think with that, we can let Dr. Hu move on to other things that you need to do in your life. Dr. Hu, I love that. Love that mug. Love that mug. It's beautiful. And your hat is amazing and your shawl is amazing. So thank you so much for your time, man. We'll see you around. speaker-1: Thank you, Lon. speaker-2: Yes, I'm going to my time machine. speaker-0: Thanks again. We'll see on the other side. speaker-2: Yes. Bye bye. speaker-1: All right, now we're coming to the end of the episode. Just a couple of topics here. We're going to remind the audience that we are running the VIO Monthly Challenge, and we're going to extend our April challenge into May. And here's what the challenge is. Anyone from the audience can make a social media post about personalized nutrition. That can be either your experience or a publication or a link to a resource, whatever you think, but highlight that it's personalized nutrition. include our podcast hashtags, so that is hashtag two PhDs on a pod in your post, and then share the URL of that post as a comment on one of our podcast episodes, or send the URL to us at podcasts at Viome.com. And then we will pick the winner based on the content and the engagement of your post. That's it. That's all you have to do. The winner will receive our premium consumer test kit, which is called the full body intelligence. speaker-0: All right, thanks, Momo. I want to also mention to our audience the topics that are coming up in the next couple of episodes. So if you remember, we discussed the life cycle of the sample and the results in the Viome user journey. So today we talked about the user collecting their sample, sending it to the Clio lab. The Clio lab does all the steps that Momo described in Quite a lot of detail. It's really fascinating stuff. And then we also had Lan join us and she explained how we take the data that comes out of the sequencer and turn that into the microbial genomes, the genes, the functions, as well as actually the human genomes, the human gene functions as well, which is what we use in generating scores. These scores are molecular pathway scores that we glean out of all of the data that comes out of the bioinformatics pipeline, and we create reference cohorts, and we create ⁓ a process by which we can understand the phenotypes that are driven by these various molecular functions that are happening through the pathways. For example, short chain fatty acid is a molecular function that happens, and it's a combination of multiple genes that are doing specific types of biochemistry that results in butyrate or other types of short-chain fatty acids. So those are the types of scores that we're going to discuss in the next episode. We're going to go into inside this ViOM scoring engine. We talk about how we understand these molecular pathways and how we validate those molecular pathways. And then the episode after that, we're to go to the next step, which is the ViOM recommendation engine. So the recommendation engine takes these molecular pathways and functions that we have gleaned from all of the data that we've analyzed so far and turns them into actionable results and recommendations for you that tells you what kinds of foods would help you because of imbalances in your pathways. What kinds of supplements can actually make a huge difference that can actually improve and even enhance the health. And that actually also results in the precision manufacturing and the precision supplements. that are part of the experience of being a Viome customer. So we'll take the rest of this loop and we'll complete it in the next couple of episodes. So that's what I am excited about sharing with all of you in the audience. And Momo and I will probably bring in a few of our colleagues and explain each of these steps in the next couple of episodes. I hope you come back and join us. speaker-1: Those are going to be super exciting. Absolutely. Translating actual molecular data into what should people eat and what should they not eat and how their physiology is working in different places, how well it's working, amazing stuff. So thank you, Guru. And that's a close. I'm Momo. speaker-0: And I'm Goro. speaker-1: Where two meHDs on a- Two PhDs on a pod is a Viome Health Science podcast production written and hosted by Viome CTO Guru Banavar and Viome CSO Momo Vujicic. Produced and edited by Lisa Shomo. Podcast art by Kevin Wu. If you enjoyed today's episode, like and subscribe for new episodes every other week. We'll see you next time.