post Podcast

Deep Biomarkers of Aging and Longevity – EP20: Polina Mamoshina (Deep Longevity)

button-play Listen To Show

Lee: Hello, Polina, and welcome as guest number 20 on Quantified Health, Wellness & Aging Podcast.

Polina: Hi, Lee. It’s my pleasure to be here.

Lee: So, first of all, where are you?

Polina: Well, I’m currently in Moscow, in Russia. So back to my hometown.

Lee: Oh, I didn’t realize that. For some reason, I thought you were in London.

Polina: No. I was living in the UK for three years when I was getting my degree from Oxford, but I left, I believe, in August last year.

Lee: It might be off topic, but I was wondering, what’s the coronavirus situation in Russia?

Polina: I mean, it’s a bit hard to comment. All of my friends and my colleagues, we self-isolated. So we’re trying not to go out. If we want to go to the office, we do this quite early. But the cases are growing, even from official stats. Yeah, it’s even worse than it was in May, in terms of numbers. So we’ll see how it goes.

Lee: Okay. And people are not wearing masks and things, in shopping, from what I’ve gathered?

Polina: Well, actually, they were in Moscow. So it’s not like I can tell for the whole country, because Russia is obviously so big. It’s a bit hard to know what’s going on out there. But in Moscow, they’re wearing and there are actually some shops, coffee shops even been closed if people are not wearing masks. I’m not talking [only] the staff. So it’s actually, I mean, the government is trying to control as much as possible. But of course, people are skeptical. But at least we don’t have any protests against coronavirus.

Lee: Okay. This is really interesting to hear because, again, it conflicts with other things I’ve heard. So it’s just amazing how different sources give you different information. I’ve been really curious about Moscow because of this level of conflicting information.

Lee: So you mentioned that you studied at Oxford. Computer science?

Polina: Right. Yes.

Lee: And what were you doing on the computer science front?

Polina: Yeah. I was part of a computational biology group that is headed by Blanca Rodriguez. She’s a professor at Oxford. And so, I did my PhD in computer science by developing biomarkers of drug responses.

Lee: Biomarkers of drug responses? That’s quite specific.

Polina: Right. Yeah. I was predicting… It was even more specific. So what I was doing, I was predicting… I was building models that can predict whether this drug is going to cause any sort of cardiovascular complications in humans. And I was doing this for all number of drugs. Those that have been approved, those that actually been withdrawn, especially due to cardiotoxicity. So that was the main theme of my thesis.

Lee: And was that something that you were interested in before you went, before you applied for that PhD? Or did that just come along by accident?

Polina: I was always interested in computational models. So I am, let’s say geneticist by training. Of course, I was trained as a biomathematician, but apparently, bioinformatics and computational biology is, surprisingly, are completely different things. And so, I was trained as a biomathematician, but I was always interested in machine learning or advanced stats, and so on. And so, when I was applying for my PhD, that’s what I wrote in my proposal, that I would like to create some models, machine learning based models that can predict drug responses or, somehow, quantify the wellness of individuals. And so it was pretty much aligned with what I was doing when I was part of Insilico, and what I’m doing right now with Deep Longevity.

Lee: And so why were you interested in machine learning? What was it that sparked your interest?

Polina: Well, it’s actually fascinated me because when I started working with data as a biomathematician, and I was mainly working with next generation sequencing, so called, NGS, I was working with a really large number of samples. And it was so hard to analyze them, but even being trained as a geneticist, it’s not like I can actually… It’s not like I was able to make any sense of the data. Not any sense, but to understand it fully. And I was always looking for some tools that help me to visualize it, visualize the data, help me to analyze it in some way. And machine learning was just perfect for it.

Polina: Think of like a genome. We have 28,000 genes there. It’s not like you can understand and know them by name, all of them. It’s impossible for human intelligence. So I believe that machine learning is just powerful tools that can help us, enable us to analyze the data in the right way.

Lee: Yeah. So you discovered it as a powerful tool. You recognized a power in it, and you were able to start wielding that power. So I guess you felt like a magician acting upon the data.

Polina: Right. Exactly [laughter]. I mean, I’m still feeling that. When I was starting actually programming, I was rather young in that, I was at school. I was always feeling really empowered with the code I was writing. Actually, my first programs, they always started with welcoming me as a creator or a master or so on. I think it’s really cool when you can actually make a machine do something for you. I mean, it’s rather empowering.

Lee: The world is going to have to watch for you, Polina. You’re going to take over it one day. You and Elon Musk and Neuralink [laughter].

Polina: Yeah. I mean, no. I’m not going to be [inaudible] if you are talking about this [laughter].

Lee: You’ll be controlling pigs with machine learning next [laughter].

Polina: I don’t know. I’m not sure I will do that [laughter].

Lee: Okay. I’d better stop joking around here. So what I want to know is, how you ended up with Insilico, and now the spinoff Deep Longevity. I think you mentioned somewhere, you were second year university. I guess, second year? I don’t know. Was that at Oxford and you attended a hackathon or something along those lines?

Polina: Right. So I got my first degree from my school state university in, as I said, in genetics. And then I was always trying to work. So I joined a lab when I was in my second year of my degree, and then I was constantly looking for some position. So I was, at some point, working in two or three labs. I mean, I was always interested in bioinformatics, and in my lab, that was my main lab that I did my coursework with, it’s not like they were actually focused on that, on bioinformatics. They had data but they had no one to analyze it, so I was the one who was doing this. So I was basically doing all major part of digging and understanding how it works by myself. And I enjoy it a lot.

Polina: But then I saw this hackathon in bioinformatics. So a competition, 48 hours, coding nonstop for biomathematicians to solve some problems. So I thought, “Why not?” I really want to join a community. I was doing it as a [inaudible] cyber community. And I said, “Yeah. It would be great to attend it.” So I went there, and apparently this hackathon was organized by Insilico Medicine. So the company that’s focused on applying machine learning to drug discovery mainly, but at a point when they were organizing hackathon, they were also focused on biomarker development and just basic data analysis and biology. And they offered me a position. So after the hackathon. So I felt, I mean, that’s great. It’s really great to find such company in Russia, in bioinformatics. So it’s a dream job.

Polina: So I started as a junior scientist, and then I end up to be a head of biomarker development when Insilico decided to separate the biomarker business, to split companies, but spin off Deep Longevity. So, of course, it was only natural, with me being from the beginning with the company, and leaving the biomarker team to actually join Deep Longevity and lead Deep Longevity. So that’s how I got here.

Lee: So that was a hackathon when you were second year in Moscow?

Polina: Yeah. I was already fourth… I was finishing my university. So it was in Moscow. Yeah.

Lee: Oh. So do you mind me asking what age you were then?

Polina: I was 22 or 21.

Lee: Yeah. That’s quite young. And was that the first time you came across biomarkers?

Polina: No. Not as a term, but no, as a term, I knew it for me. Because biomarker, in general, is such a broad term. You can call almost anything a biomarker. I mean, something that reflects a state or a pathological disease state, a pathological state in the human body. So, albumin levels, those are also biomarkers.

Polina: But that was probably the first time when I encountered new types of machine learning models. Because machine learning, in general, it’s just statistics. So I was applying some of the things in my university when I was visualizing data and doing the data analysis. But only when I… Then, at hackathon, I learn about the models, and so on.

Lee: Well, thank you very much, Polina, for that personal introduction and letting me ask. I’m going to read… I just looked at the description of Deep Longevity on LinkedIn. Just let me read it out.

Lee: Deep Longevity is developing explainable artificial intelligence systems to track the rate of aging at the molecular, cellular, tissue, organ, system, physiological, and psychological levels. It is also developing systems for the emerging field of longevity medicine, enabling physicians to make better decisions on interventions that may slow down, or reverse the aging processes. Deep Longevity has developed Longevity-as-a-Service solution to integrate multiple deep biomarkers of aging, dubbed, quote, deep aging clocks, to provide a universal multifactorial measure of human biological age.

Lee: And here it says, originally incubated by Insilico Medicine, Deep Longevity started its independent journey in 2020 after securing a round of funding from the most credible venture capitalists specializing in biotechnology, longevity, and artificial intelligence. Deep Longevity established a research partnership with one of the most prominent longevity organizations, Human Longevity, Inc. to provide a range of aging clocks to the network of advanced physicians and researchers.

Lee: So I think having that as some kind of, at least, company introduction. The other thing I should mention is, Deep Longevity only recently came out with stealth, and then had this major partnership with Human Longevity Incorporated. And then, just very recently, 2nd of September, they were fully acquired by Regent Pacific Group. So it was quite a quick set of announcements, after coming out with stealth.

Polina: Yeah. We’ve been in stealth for a year at that point, and, as I said, because it was a part of Insilico, the technology was already mature. So one of the assets that we have is this Young.AI app. So we’ve been building it for several years, and now it’s a part of Deep Longevity. So, I guess, that was the main reason why we were able now to secure those deals pretty fast.

Lee: I think what we should do for listeners, and I do hope to have the CEO of… Alex, who’s agreed to be on in the near future. But could you provide just a basic introduction to Deep Longevity?

Polina: Right. So, as you said, Deep Longevity, it’s a company that is focused on developing aging clocks. So, those tools that can predict the rate of biological aging of individuals. And so, predict mortality, their life expectancy, and so on and so forth. And the main idea that’s with those two is we can actually enable clinics, enable consumer companies, to test their efficacy of their interventions, also to test the effects of lifestyle changes on humans.

Polina: Because we can actually quantify them. We can quantify how old you will be if you quit smoking, or start smoking. We can ultimately, with those tools, run clinical trials on longevity interventions. That will be the ultimate goal for us.

Lee: What will be your ultimate goal?

Polina: To run a clinical trial on longevity intervention. I mean, it’s not ultimate goal for the company. So to say, the company actually has really big plans. I know it’s silly to say, probably for a startup, but, again, what we want to do, ultimately, we want to create an ecosystem. And we believe having biomarkers of aging as tools we can track so many things that would benefit us a lot. There’s actually something that can build an ecosystem, and something that we can use to build it. Yeah.

Lee: Are you able… Yeah, I meant, able, in terms of, from a business privacy, confidentiality perspective. Are you able to share any details of that ecosystem?

Polina: I mean, probably not yet. So we haven’t announced that yet, and we are still in the middle of closing the deal with Regent Pacific. So the company that acquired us. So at this point, I don’t think we can announce all the details. All I can tell is we are right now working with our clinic partners and aiming to assign more and more clinics to work with. And of course, launching the app. So we really want Young.AI to be big and have a lot of users.

Lee: So, you said Young.AI. And Young.AI is your app. I was a little confused because I always knew Young.AI as the web URL where you could entire blood analytes and get a biological age prediction. And then, the other week, at the press release and demonstration, suddenly Young.AI was an app.

Polina: Right. Yeah. So those are actually two separate things. So we have Aging.AI. That’s a website where you can enter blood parameters.

Lee: Yeah. Sorry.

Polina: Yeah.

Lee: Yeah. I got so confused, even over the last few years. I totally forgot again. I used to get them backwards. You had the URLs, Young.AI, and Aging.AI. And it’s Aging.AI where I entered the blood analytes.

Polina: Right. Exactly.

Lee: But Young.AI also used to be a website of some kind?

Polina: Yeah. I mean, it’s a website. It’s a web application. I mean, it’s actually system of two. We have a mobile app in development, and a web app. So, the web app, you can use with your browser, open just as a website, log in there and enter information, and so on. And the mobile app is something that complements the web app.

Polina: But Aging.AI is still out there. It’s functional and it’s open. It requires no registration. And some people prefer to use it. But Young.AI obviously is way more powerful in terms of data visualization, what you can synchronize with it, and so on.

Lee: So the app is… You’re calling it Young.AI. Could you give a little tour… We shouldn’t say tour, because people are listeners and not viewers. But could you give a description of the app?

Polina: Yeah. So Young.AI is a companion app. What we’re aiming for is to have something that will allow our users to learn their aging rates on a daily basis and adjust from it. I mean, if they want to. And that’s why we decided that we want to build a mobile app. So there’s something in their pockets, on their phones that they can check from time to time. And the app will help them to get younger.

Polina: And we’ll also have a web app, which obviously is way more powerful than mobile app, in terms of how it can visualize data, what kind of data types you could upload there. And so on. But ultimately, a system of two. So they both synchronize. If you have account in one, you automatically have an account in another one. And if you upload the data on the web, you will see it in your mobile and the other way round.

Polina: The basic idea behind Young.AI, it’s a system that allows you to upload your data, like your photo or synchronize your Fitbit, or Apple iWatch. And track your progress over time, see what makes you younger or older, and adjust from it. And in terms of adjustments, we even provide some suggestions along your, say, your big points. For example, if you have some issues with your sleep, like irregular sleep patterns, or you sleep too much or you sleep too little, or you have elevated heart rate during sleep, suggesting that you probably have higher stress levels, and so on and so forth, we can actually pick this up and suggest you to pay attention to it and what you can actually do to change it.

Polina: And the basic idea from it, we can actually affect aging rates by adjusting those small things in behavior and lifestyle. Probably easy things to do, but still, I believe that they will make a big impact on aging rates of individuals.

Lee: I think our age is… And the terminology may not be right here, but I’m sure you’ll get the sentiment, that our age is probably the greatest asset that we have, because it’s a proxy for health. I now definitely say that health is our greatest asset.

Polina: Yeah.

Lee: And… Okay. Yeah. Thanks. And so what this app is aiming to do is to slow the rate of aging, and therefore what you’re saying is, “Protect an asset. Protect your greatest asset.”

Polina: Well, ultimately, yes. We want people to live longer and healthy life and the app is aiming to do this, to take control of your aging process. Help you to track and understand what is going on there. And ultimately, protect your, as you’ve said, the major asset. Yeah.

Lee: Yeah. So in a sense, it’s your most precious asset. And the app aims to protect it, stroke, improve it, then it’s a logical follow on that there is money in this field.

Polina: Yeah. For sure.

Lee: And back in 2015, I could see, what we call, this market, without giving lots of words to it, this market emerging. And I thought it would begin, sort of 2019. It’s been… Okay. A touch delayed by a year. But this year, I definitely see the beginning of the hockey stick. So you’ve been in this long enough, also. Do you also see that this year is what I’ll call, a launch year?

Polina: Yes. I totally agree with you. That we will see right now. And probably that why it was so easy for us to fundraise, because, ultimately, we thought, “Okay. COVID is there. An economical crisis is probably going to hit.” I mean, it’s already happening. So with all that, it would be really hard to fundraise. But then, no. Again, because we can actually see right now that the market, the longevity market is emerging. So there are players like HLI, other clinics like Calico, companies who do drug discovery and development, like resTORbio, or [inaudible]. All of them. They are creating ecosystem and there will be more and more companies emerging like that.

Polina: And I think the main thing that actually enabled this is biomarkers of aging, because the longevity field has been out there for quite a while, and have been feeding us promises for quite a while. And there are still a lot of products out there that are mainly like snake oils. They do nothing. But now, we can actually track effectiveness of therapies. So now we can actually quantify how effective they are in terms of reversing aging. Maybe not reversing, but slowing it down. Ultimately, of course, reversing. I think that because biomarkers of aging, it’s a completely different story now. And that’s why the field is booming.

Lee: Yeah. So you mentioned a few things there. I think the COVID situation is actually going to accelerate the industry rather than delay it because I notice it’s almost like the public has discovered death and morbidity all of a sudden. And it seems to have put the idea of health protection and aging far more into the public consciousness.

Polina: Right. Totally agree. That will we see, because it’s a geriatric disease, so the age, ultimately, chronological age, it’s one of the most important factors of predicting you get any complications from COVID. And, again, the public starting to pay a lot of attention. It actually will be really interesting to see, once we will be able to manage COVID, and I believe that we will do in the next couple of years, maybe a year, how the public perception will change, or how fast it’s actually going back to normal, in terms of thinking around health.

So there’s something in their pockets, on their phones that they can check from time to time. And the app will help them to get younger.  
Polina Mamoshina

Lee: I definitely think that we’ll come out the other side with health definitely more at the forefront of public consciousness.

Lee: So back in 2015, when I had the realization that, hey, look, there’s a new multi trillion dollar market emerging, and it’s a market I want to be involved in, because healthcare… Well, I’ll just jump in the deep end here and give a personal opinion, which is that the greatest danger to your health can be healthcare today. And it’s due to misaligned incentives and paradigms which are exceptionally out of date. And so I wanted to help towards… One, I wanted to part of this emerging multi trillion industry but I also wanted to facilitate the paradigm shift. And we can call that at the moment, slowing aging and reversing aging, reducing chronic disease risk, et cetera. And so, I wasn’t sure where to study or where to put my attention because there are so many areas to it.

Lee: So I decided the nucleus or the core, the juncture, the best place to place myself would be at the quantification of health, wellness, and aging. And so, I’m deeply interested in the fact that you’ve chosen the quantification side on Deep Longevity. And could you introduce aging clocks?

Polina: Right, so aging clocks, they are those biomarkers of aging that return age as a number. So, predicted age and biological age is a number. And there are a number of those that has been proposed recently. It all started in 2013, with the publication of epigenetic clocks. So Steve Horvath thrust it forward, was one of the groups. He was the first one to publish, but also there was another group that published, actually, around the same time. So it was tricky with those peer reviewed journals.

Lee: What was that group?

Polina: So there was another publication that is the first author is Hannum, and it was published almost the same time with the publication that Steve Horvath did in genome biology in 2013. But Steve, of course, was the only author there, and the other publication was for multiple authors. I mean, I can send you the link, if you like.

Lee: Yeah. Sure. And I’ll add it to the show notes,

Polina: Right. So, again, but Professor Steve Horvath is considered to be a father of the whole field, with his first approach. And since then, those been proposed in transcriptomics, gene expression data. Also even in telomere length. So there are a number of models being proposed since then. And that was mainly driven by the availability of data. So right now, we have all the data to predict, to try to design those clocks, that predict biological age.

Polina: But then, in 2016, we were the first one who decided to use a completely different technology to build those clocks. The clocks, the epigenetic clocks that were developed in 2013, they were built using so called shallow machine learning techniques. So really standard, basic machine learning. While we decided that we can use a superior technology that this… It was at the point quite popular, and still really popular, deep learning. So, in 2013, I did learning based model, actually, won the ImageNet competition. It was a competition in an image recognition field, and then neural networks actually show that they are more accurately predicting and classifying images or recognizing them than human experts trained to do so. So that was a big deal at that time.

Polina: And we the company decided that images, they’re really complex in terms of their structure, and complex in terms of their analysis, but gene expression, omics data in general, biological data, is even more complex. So why don’t we try to apply those deep neural networks to analyze this biomedical data? So we did it, and apparently, they are also superior to the standard machine learning models in aging clocks. So they’re way more accurate, and sometimes they are the only way that you can actually apply to build a model for some of the data types.

Polina: Mainly, again, for imaging. So there are several models that are being proposed for MRI based clocks, also for X-ray based clocks. So lots around the image based that obviously working better with deep neural networks. But also, omics is apparently better analyzed by deep neural networks than just standard.

Lee: Why is it better analyzed? Because I can see why, on the vision side, if you want to guesstimate someone’s age based upon photographs, I can definitely see why neural networks would be of benefit. But why would it be of benefit in the plain omics side?

Polina: Right, so, on the image side, it is because, well they usually have higher resolution. So if you think of images as like a collection of numbers that encode the colors, and so on and so forth, they’re really high resolution. And deep neural networks are really good with working with those data types. But one that worked with image as well, those are a bit of different architecture. So that was the so called [inaudible] neural nets. So that can actually deduce the [inaudible] a lot. And one of the benefits they also have is that they kind of use those filters that walk in through the image, so they can pick up features or things, particular characteristic or image, regardless of the position of this characteristic or image. So, regardless where your cat is, the neural network can actually recognize it’s an image with a cat. So that’s why that’s so powerful.

Polina: And if we think of gene expression data, for example, it’s also sort of an image. So it’s a snapshot of total… I mean, not totality of proteins expressed, if you think of mRNA analysis, and it’s just a snapshot of the state because the gene expression is rather dynamic.

Lee: Yeah. I imagined this huge “analog form” when you were talking there.

Polina: Right. So you are taking a picture of it at a particular point of time. Again, it’s also highly dimensional, in terms of the number of input parameters that you usually analyze… starting to analyze 3000 or 5000 or all the way to 8000 genes, human genome. And if you think of neural network as powerful tools, what they also can do, they basically are able to capture really highly non linear dependencies between…

Lee: That’s a key. And for the sake of listeners, can you explain what you mean by that?

Polina: Right. So-

Lee: The high non linear dependency.

Polina: Right. So, I mean, if we think of blood markers, for example, and we analyze the albumin levels in blood over time, and trying to capture any dependencies between their chronological age and level of albumin in humans, we will see that it’s actually declining with age. So there is a linear trend there, and we can express this dependency as a linear equation of some sort.

Lee: And same with glucose.

Polina: Yeah. Same for glucose. Yeah, glucose tends to increase, cholesterol tends to increase, but again, it’s a linear dependency there. And also, the models that been used, like the shallow machine learning models, they actually are the one that are trying to capture those linear dependencies, find those coefficients that actually express in their, convert in their input levels of blood parameters to the target level. So it’s ultimately like one equation, with multiple coefficients.

Polina: And with deep neural networks, it’s actually multiple equations, and they also stack together, and if before… You’re not actually converting input parameters directly to your target as age, you are doing this in multiple levels, and you have multiple transformation before that. On the hidden layers, the so called hidden layers, you actually don’t analyze them. You only analyze the one that you see or have access to the one that go on the input, and then something that you receive as a target, for example, as in age. And then, hidden layers, let’s say all the magic, are accurate, so all those non linear transformations, multiple transformations.

Polina: And because of that, a deep neural network can actually work with really complex data types because usually, even if you have those obvious linear relations within, like albumin level and age, it doesn’t mean that if you have multiple parameters, albumin, glucose, cholesterol, that you can actually express a dependency in a linear form. It’s not that simple. And even for albumin-

Lee: Not that simple?

Polina: Yeah. It’s not that simple. And even for albumin, if someone will give you an albumin level of someone or even myself, it’s not like you can actually guess how old the person is. But having a combination of parameters and the linear models that would work, would solve the problem.

Lee: There’s so many things I want to ask you there. I really appreciate that. So the first thing that comes to mind is, has no other group taken this deep learning approach?

Polina: Yeah. There are other groups that do this for other data types, including imaging. But deep learning-

Lee: I understand for imaging, but for aging clocks?

Polina: Right. I mean, not that we are aware of. Of the…

Lee: I’m really surprised. So, I…

Lee: …I’ll be honest with you here. I assumed that Alex or Insilico, now Deep Longevity, my assumption was they were just throwing in the word, deep learning, as a marketing term.

Polina: No.

Lee: So now, listening to you, I completely get this is quite a very unique approach. And it contrasts with the approach of Morgan Levine.

Polina: Right. Yeah. It’s also a bit different in terms of how we design clocks. But coming back to your question on deep neural networks, again, I mean, yeah, they are powerful, but there are setbacks, of course. Because you have, let’s say, so many equations inside of it. You have so many parameters that you have to fit. In a linear model, you just know, try to find an optimal coefficient that will give you best results, like (inaudible) model. And you can understand that to get best results, if you, for example, you introduce some error in there, and penalize them, although if it producing a lot of errors in the response.

Polina: But with neural networks, you have so many parameters. It’s millions of parameters, not even thousands. And they’re really hard to train. It will take time. And they are so hard to train that they actually have to have, let’s say, special computers for that. It’s not like you can train them on CPU straightforward. You usually use GPU for that, so graphic processing units. The one that actually analyzes images, but they are really apparently really good in working with [inaudible] of data. And because of GPU, now we can actually train deep networks.

Polina: Because the whole theory behind deep neural networks and positrons have been out there for a really long time. So most of those ideas, they came from ’80s. But now, because we have computer power to train them, we can do that. It’s not only on the computer part, but also algorithms. Before, it was impossible to train so many parameters, but now we have a back propagation, other algorithms that allow us to train them.

Polina: So a lot of things actually happened to enable us to apply this technology to manipulate data analysis. Plus they are exceptionally data hungry, so if you have not that many samples, it’s not going to work. And also, it’s a bit hard to find people who can do this sort of analysis. It’s more popular now. And there are degrees right now in machine learning, but it wasn’t like that a couple of years ago, even.

Lee: I remember in 1994, trying to program neural networks. And apart from some limited success, maybe some medium success on images, I thought this is just complete rubbish.

Polina: Right. Yeah [laughter]. I mean, yeah, because the computer power is not there, algorythms not there, data not there yet. And yeah, it was rubbish, nothing. I mean, not rubbish, but more like a theory, not…

Lee: Yeah, exactly [laughter]. So I was wondering why we were doing it, and why so much effort was going into it. And then, a number of years ago, with machine learning, that seemed just way more exciting and efficacious, and I totally put deep learning to the back of my mind. And now you mention this as an approach to quantification of aging.

Polina: Exactly. And because of that, it’s probably hard to find people who actually trying to solve, and especially it’s hard in biomedical field, because it’s not only you have to know how to train neural networks, you also have to understand the data, how to process it, what kind of issues are there, how you design your experiment, in terms of exploiting by test train set, and so and so forth. So the multiple things you have to know to do this. And that’s probably why we don’t still have that much competition in this space.

Polina: But even though the accuracy of deep neural networks is superior to other models, other models are still good enough. So for some of the companies for some applications, they don’t prefer not to go this road. Getting people trained, highly skilled to do this, getting the equipment to do this, and so on. They’re kind of okay with other approaches. And mainly from business point, it’s also a viable model. But we believe that deep neural networks are better, and it’s something you should invest in if you want to be competitive in the field.

Lee: Yeah. So I’m really interested in this two schools of thought going on here. So with the Morgan Levine stuff… Well, I used Aging.AI so I could let me… Did I get it round the right way? Yes.

Polina: Yes.

Lee: Because I get it confused with Young.AI. I used Aging.AI and I would go get a cheap blood panel, and occasionally fill in analytes and just see what [biological age] score. So you’re saying that was neural network based but I think, if I remember right, the R value was 0.8 on Aging.AI.

Polina: Yeah. So for… It depends on the number of parameters you’re entering, because for 41 based panel, for 41 based blood parameters based clock, we had R value close to nine. So it was nine something.

Lee: Which clock is close to an R value of nine?

Polina: It’s for 41 parameter based. So…

Lee: I actually did that once, by the way. I went to the blood lab and I took all 41. So you have one which is something like nine or 10 markers, which I tend to do nowadays. And it’s one which is an R value of 0.8?

Polina: Yeah. I think it’s even less. 0.76, something like that.

Lee: Oh, okay. And then, Morgan Levine, you know who’s been a postdoc of Steve Horvath, in 2018, she published PhenoAge, and that was an R of 0.94. And it uses nine circulating biomarkers. And so it’s clearly very good. And with Elysium Health, they’re now selling that biological age score, and the intervention of nicotinamide riboside from Elysium. And what you’re clearly saying is you’re taking a… So that’s a machine learning approach, and what you’re saying is you sincerely believe, though, that this deep learning approach will yield better results now or in the long term.

Polina: Right. So there is an issue there with accuracy. Again, so there are two schools of thoughts here, two ways of approaching the design of a aging clock. One is that are you suggest that there are some markers in a body, measurable, or let’s say, blood levels, or I don’t know, walking speed rate, whatever, that you can connect to age. So your input parameters will be some biomarkers in the body you measure and output will be the biological age. But because you don’t know the biological age, so it’s something that you cannot actually know for sure, so there’s no ground truth there, you would assume that your chronological age should be close to your biological age and that, what we assume, in healthy individuals. I mean that’s the reason we have mortality tables and that’s the reason we can predict the lifespan, right now, for healthy individuals. And so on, and so forth.

Polina: And you design those models by minimizing the error between the age you predict with a model, with input parameter given, and the chronological age of healthy individuals. Of course, those models are never perfect. Even though you try to minimize the error, there’s still some reasonable error rate still and you see some variation, and you believe that it would assume that this natural variation, because, again, even though we can predict lifespan, still, different humans, they have different lifespans, different disease incidents, and so on, and so forth.

Polina: And another approach would be to include the chronological age as one of the big parameters and assume that your biological age, it’s some kind of adjustment to your chronological age. Again, so you will assume that in your equation, are you not only have blood parameters or some other markers as input, you also have chronological age. But what you will put on the other side of your target? Because you don’t have biological age. It’s an unknown ground truth. You have to put something there. So some of the models that actually prove the chronological age because you are again making the assumption that the biological age of healthy individuals is kind of close to their chronological age. You would assume there’ll still be variation, that’s why models are never perfect. So you go for it.

Polina: But if you think of this from a machine learning perspective, it’s not going to work because you have your chronological age as input, and you have it as output. And the perfect model will just minimize any coefficients before other markers and equalize the chronological age to chronological age, regardless what you have out there.

Polina: So the training there is just not obvious. When will you start? Because, I mean, you can start from model that is not very good at predicting, and model will try to go in a direction by minimizing the error, and so adjust the coefficients. And of course, you can stop at some point, not allowing it to be the perfect model, but ultimately it will still go there. So the [inaudible] is just out there, because again, you just equalize chronological age to chronological age. And that’s why we’re not in favor of those approaches because it’s impossible to understand when you will start.

Polina: So now with those groups that still use this chronological age as input parameters or some adjustment that they use it… I mean, you can call it a trick. But of course it’s a valid assumption that biological age is like modification of chronological age, some adjustment, but still, you can call it a trick because it’s way easier for us to predict age when we have chronological age.

Lee: Yeah. So it’s a covariate, chronological age.

Polina: Exactly. And then they are trying to introduce mortality as one of the things that you can predict, so you’re not predicting chronological age directly but you’ve predicted mortality. And that’s how their GrimAge, for example, works now. So…

Lee: Yeah. Steve Horvath. Was that 2018, also?

Polina: Yes. Yes. Yes. It was 2018.

Lee: But what did he call the mortality part? Was that, time to live, or time to death? Yeah, it something like time to-

Polina: Yes. Time to death, I think. Yeah.

Lee: Quite a name, of GrimAge. Time to death.

Polina: I mean, Steve is rather good at marketing. I mean, you should agree on that. Everyone knows GrimAge…

Lee: The hacker community loved that.

Polina: For sure [laughter]. And then you still have a problem with chronological age. You still have it, and it’s, from machine learning point of view, you really don’t know where to stop because when you train a model, it’s still going… You still have situations and trying to minimize other coefficients to optimize…

Lee: Sounds a lot of fun.

Polina: For sure [laughter]. It’s actually, I mean, I love it because it’s allow you to think, if we are not… bit of side topic. If we compare wildlife biologists who work in a molecular lab doing some experiments and conducting some reactions, and so on, they still at least have time to think, because usually those reactions are time consuming. So PCRs, for example, with running one and a half hour time and you can actually spend a whole day preparing the data for it, like you extract the DNA, and so on.

Polina: But for now, for computational biologists or computer scientists, you can actually get results of your experiments in seconds. And it’s sorta frustrating, you really don’t have to time to think about science and so on. So for neural networks to be different, because, again, you can spend a whole day training them, but it does require your attention, but the time, you’re kind of in the same position as a MATLAB biologist, to think, to make hypothesis, to read literature. So it’s a good thing.

Lee: Or, on social media, right?

Polina: Yeah. You can do this too.

Lee: You’re making it sound like you’re always focused on work. It reminds me. I laughed at your… Today, I looked at your Twitter bio, and it’s, “Geneticist by training, bioinformatician by trade.” So I liked that humor.

Polina: Right [laughter]. I’m not using Twitter much. But yeah.

Lee: This GrimAge and the time to live talk, you mentioned longevity clinics as a potential customer of Deep Longevity. I guess that would be through the human longevity partnership, but maybe not necessarily. I’m not sure how you’re going to structure it. But also life insurance must be a big market.

Polina: We hope so, because if we think of this from a perspective of having people living longer, not getting any disease, that should ultimately benefit both sides, because then life insurance companies are not going to pay any incidents, and then a human… I mean, humans are going to live longer and happier and…

Lee: And pay more monthly subscriptions.

Polina: Oh, for sure. Or maybe less because I think the proper way to design it is actually allow people who take care of themselves to pay less. It’s only fair because they have lower mortality rates and so on.

Lee: People try to achieve that with cryptocoins. Making health coins and you earn more coins based upon your living. But it was very poorly implemented.

Polina: Right. Yeah there is always a gray side of it. Gray side of it. So it’s a bit hard to control properly. A lot of things can go wrong, yeah, again if you implement this poorly.

Polina: But if you think of how the life insurance should approach it is that it’s more like of a marketing tool for them as well. Because when you ultimately start learning about your risks, and that was a study run by Swiss Re, on their genetic testing. So they showed that after doing genetic tests, people are more prone to purchasing life insurance regardless of them actually finding out-

Lee: So it’s like an onboarding solution?

Polina: Exactly. Yeah. So regardless of them finding out if they have any sort of chances of increased risk of any diseases, they’re still willing to purchase life insurance three times more. And that’s ultimately a good thing for life insurance companies.

Lee: Plus they have a connection with your personal data, and they can provide dashboards, even if it’s white labeled, dashboards that have been white labeled to them. And it’s nice to have a dashboard with a product now.

Polina: True. And there is a company called yousurance.

Lee: Yeah. I’m aware of them.

Polina: Yeah, right. They use epigenetic data to actually calculate your premium.

Lee: Talking of data… Sorry, did I cut you off there, Polina?

Polina: I was about to ask, do you know how well they are doing, if you know them?

Lee: I’m just aware of them. I don’t know anything more. There’s a couple others, actually. But I also know that that market is also set to take off over the coming years. So I see a lot of stars lining up.

Lee: But talking of data, with Morgan Levine’s PhenoAge, the 2018 paper, given the nine analytes, and also it’s a composite marker, so there’s the methylation also getting measured if you want. They used NHANES III data for training their model. I haven’t heard any mention of what datasets have been used by Deep Longevity or Aging.AI.

Polina: We used the same dataset for validation but our main source of data is our cooperation with universities and clinics. So we have a really comprehensive internal database that we actually augmented also. Because now, what we are trying to do is to work with as many markers as possible but also as little as possible. Because when you start work in the consumer space, you realize that there are not that many people who are willing to purchase a thousand dollar blood panel once a year, even, to do the testing and to analyze all of those blood parameters.

Polina: And to do this properly, we artificially augmented the data and created a lot of artificial samples with the missing values, so now we can work with the missing values as well. And that help us work with as little blood parameters as possible. And that’s why, for the previous versions of Aging.AI, we had those free panels. But now for Young.AI, we don’t have a panel, because Young.AI is able to work with a different set of parameters.

Lee: If I look at Aging.AI at the moment… Well, if I look at PhenoAge, its nine blood markers, you’ve got the really cheap analytes, albumin, creatinine, glucose, you’ve got the mean corpuscular volume and MCV, the red blood cell distribution, RDW, your alkaline phosphatase, white blood cells, and your C-reactive protein.

Lee: If I jump over now to Aging.AI and I click the one that seems to require the least… Let’s have a look here. Yeah. 19 input markers. They’re super cheap for me to get here and fill out. Yeah, so you’ve got extras on top of this nine, like potassium and your calcium. Strange that calcium is there. First question, has this got you interested in blood biochemistry?

Polina: Yeah. For sure. Because now we are having a model that can predict age based on the blood biochemistry levels, or cell counts, we can actually design a model that can suggest what the optimal levels are for this particular individual to get younger. And for this, of course, you need to know what is going on there, what kind of interventions or conditions impact those parameters, of those blood parameters, their levels, and whether there are any ways of changing them.

Lee: I find it fascinating because when I first came into… What, again, I’ll call, this industry, in quotes, and I spoke with many physicians, first of all the attitude was you never measure something unless you know you can take action on it. So it was very conservative paradigm when it comes to data. I have to watch what I say here, but I definitely dealt with physicians at a very close level for a number of years, and they showed me what they were doing and they were just looking up reference ranges in a pathology flow chart… If it’s this… And it was one number pretty much applied to everybody. The same numbers. It was just a lookup for a pathology diagnostics.

Lee: But then I began doing my own experiments in the labs on myself for years, and it was like, damn, I can change one value and increase my lifespan by potentially 10 years. Let’s not look at this as a disease diagnostic bucket, of putting in a disease diagnostic bucket, but let’s look at how I can alter these values with supplements, with food, and so forth, and actually see predicted lower morbidity, increased longevity, and so forth. So I just got started getting super excited about blood chemistry.

Lee: And so maybe you’d be so kind as just to say a few words about the markers used in Aging.AI on the blood chemistry front. Maybe let a listener know why is albumin there, or bilirubin. Just a little bit about blood chemistry, if you could, Polina?

Polina: So for blood chemistry, when we were designing the set of markers that we were planning to use we’ve basically chosen the one that was the most common one. And the reason for that because we want the approach to be really massive, so it has to be really cheap.

Lee: By most common, do you mean that it’s been measured for decades and in scientific literature, and super cheap because every lab has them?

Polina: Exactly. And it’s measured like part of a regular testing in most of the hospitals all over the world. So it’s universal markers that measure everywhere, like glucose. Most of us, we already have it, and actually multiple of it.

Lee: Can I just add a side note in there with you? That do you think that blood chemistry, standard blood chemistry, the super cheap stuff that gets measured everywhere, is untapped, in terms of intelligence?

Polina: Do you mean that it’s good enough?

Lee: People will only look at one or two values but once you start feeding these super cheap markers into machines, that has not been tapped yet, what you can derive out of those markers by using machines.

Polina: Yeah. I was saying, that’s exactly why we are focusing on optimal blood ranges rather than reference ranges. Because, as you correctly said, that reference ranges are designed to detect any pathologies, like really extreme conditions. So in some cases, it could be… I mean, not too late to reverse, but still damage in there. It’s a sign of something bad coming. Of course, it’s up to a doctor to do any diagnostic work based on blood values, but still, it’s something already unhealthy. But if you think of it, there is also an optimal range, as you, again, correctly said, that you can reach with your blood markers to get even healthier or to live longer.

Polina: And when someone is looking at data, what they usually do, they tend to pick up one or two blood parameters that are, let’s say, abnormal. They are outside of the reference range. They will be the one that will be…

Lee: Yeah. Two standard deviations.

Polina: Exactly. They will be the one that will be highlighted by the lab test providers? , They are the one that doctors will focus on. And I think that’s a shame. We also have others, and you have to analyze them and also analyzing the trend in your blood markers that can be also really powerful, because even though-

Lee: So it has to become the case that we end up with optimization clinics, or we’ll call them longevity clinics, or both. And so I got excited when I saw Deep Longevity launch. Because we need this quantification to power this industry transition so we can start having money poured into increased age and optimized health instead of always reactively waiting on people to get sick and putting them in a disease bucket through quantification.

Polina: Exactly. Yeah. That’s the whole idea.

Lee: And I see the time running out, so I’m getting a little bit of a panic here that there is just a couple more questions that I wanted to throw at you.

Lee: So, traditionally, talking the last five years or so, maybe since Steve’s paper in 2013, biological age has been a single number but in the last couple of years, it seems that biological age is splitting out to subsystems. So you’ll have a heart health, a brain health, and so forth. And so, you, or Deep Longevity offer multiple aging clocks, so I guess you plan multiple biological age measurements of maybe of different bodily subsystems and not just one complete number. Do you want to say anything there?

Polina: Yeah. Exactly. So that’s correct. We don’t believe there is a one universal marker that can be used for completely different tissues, organs and systems in the body. So that’s why we are focused on a really broad range of them, starting from really simple blood tests or photos of facial images, and going to metabolomics and microbiome data, and so on. So it’s really hard core omics data.

Polina: And, of course, ultimate goal is to find the fastest ticking clock in your body, the one that is actually going to say… I mean, not fail you first, but find the one that is accelerated and try to tweak it.

Lee: Yeah. So someone could have a heart age which is well in excess of their chronological age?

Polina: Exactly.

Lee: If you could get access to that tissue, or a measure. We all seem to have a preponderance towards a certain type of accelerated aging. I think it was AgeoTypes… And now I wish I had more caffeine, because it slips from my mind. Michael Snyder.

Polina: Yeah.

Lee: I don’t know if you’ve heard of… Oh, you have heard of that. And so you gave a nice tour to the media… I don’t know, it was like 10 days ago or something like that, where you unveiled the app, and the app could begin super cheap, the Young.AI app, and it could take your biological age just taking a selfie.

Polina: Yeah. That’s the whole idea. But that’s the only way for us to make it massive. I mean, we both believe in quantification and rigorous analysis of data, and continuous monitoring. But most of the people, they don’t. And I guess it would be a really tedious task for us to enroll a lot of people with the Young.AI, but ultimately, that what we want to do, we want to be really popular with millions of users and those millions of users then will pay attention to the aging field, to the longevity biotechnology.

Lee: It’s quite clever to let people do a biological age test using a selfie. Obviously, the accuracy doesn’t meet blood based, and so forth. But because it draws people in at no cost.

Polina: Exactly. Yes.

Lee: And then you can upgrade to different levels, and, I guess, money, the more you spend, the more you tend to get an accurate score, I would say. Which brings me on one of the last couple of questions is, it seemed to me, and please correct me if I’m wrong, that you have a preference towards transcriptomics and proteomic aging clocks.

Polina: Right. Because we believe they are the most actionable one, the one that you can actually target with interventions. Because even for epigenetics, it’s a bit harder to change, to adjust.

Lee: Yeah, because you see, on the epigenetics side, you don’t know if it takes six months or two years to have an effect on histones and so forth.

Polina: Right. Unless it’s something that directly affects the levels of methylation, but it can have a lot of side effects, because those methylations, they control all the processes in the body. They are specific in some way but also really general. So if you change epigenetic profiles, you can actually have, potentially, a lot of side effects because of that. Because you’re, again…

Lee: Yeah. It’s complicated because those epigenetic changes are a sign of healthy aging.

Polina: Right. Exactly.

Lee: So that’s quite a lot to get your mind around. And the thing on the epigenetic front, you don’t know if it’s a cause of aging or if it’s a passenger of aging. But I don’t know about any transcriptomic and proteomic front.

Polina: Yeah. You would assume that… I mean, because it’s different, so you can be more specific here, especially you think of certain pathways you’re trying to intervene with. Certainly signaling pathways that control… They can be, also, again, a master pathway that control multiple processes but also they can be specific.

Polina: And you already know what kind of drugs target those, like mTOR, rapamycin, targeting the mTOR pathway. Again, it also has some broad effects that are… what kind of subunit of them to [inaudible] it’s targeting, inhibiting, but still at least it’s let’s say, more controllable in a way. So it’s more like surgical intervention there when you’re thinking of pathways. Because epigenetics is ultimately on top of it. And the higher you get in say, a hierarchy of regulatory signaling, the broader effect you will have if intervene on top rather than if you intervene on one of the branches that are really specific in their function in the body.

Polina: And both transcriptome and proteome, both of those snapshot are illustrative of the state of those signaling networks. And that’s why we believe they are more powerful, are more actionable, in terms of biological age tracking. Of course, they should be less accurate, but then there’s always a downside of it, as you said, we really don’t know whether if we are going to make an intervention on an epigenetic level, when we’re going to see an effect. And the animal experiments suggest that while it can be rather long before you see any signs on the acceleration of the aging rate.

Polina: So there was a really nice study from [inaudible] lab, from Harvard. What they did, they measured aging rates based on methylation in mice and they did it for mice for some of the Ames dwarf mice that are known to have extended lifespans, and they’re also for mice and caloric restriction. And we know that caloric restriction extends mice lifespan up to 40%, depending on the strain. But still, it’s something that works.

Polina: And what they showed that you actually have to put them one year, at least, of caloric restriction before you see an effect on their methylation based predictive biological age. So it’s one third of our life. It’s not something that you can really easily translate into human practice, if you think of it. And that’s why it’s a bit tricky with methylation. So it’s like, yes, you can predict age accurately, but then, first, why do you need to predict chronological age so accurately? Because you already have it. You have a passport. You know it. And also, because you can predict age so accurately because it’s rather stable, and so, it’s harder to reverse it, harder to change it.

Lee: It’d be nice one day, when you, for example, move location, you can see whether that move to a new city or country accelerates or decelerates your aging. For example, the air pollution accelerates your aging.

Lee: It would be nice to see that. Kind of like a Google Maps overlay.

Lee: Or, in 2015, when I said this, people laughed or gave me a funny look. I said, “When it comes to dating, the future of dating, well, I want to see their time to death. I want to see the other… If I’m planning long term, I want to see their morbidity predictions.” Don’t you think that that will become a normal part of social interaction in life, just like blood pressure is?

Polina: Right. It could be. I think-

Lee: But we do that the moment when we look at people. That’s why we like people with healthy looking skin.

Polina: Exactly. And bioevolutionarily, we’re predisposed to picking partners that are going to have the same lifespan to one that we are going to have. Because there was a really nice study, I think it was run by Ancestry… You know this company?

Lee: Yes, I do.

Polina: Exactly. So what they did, they analyzed cousins of genomes with the connection to the life expectancy, and they showed that, regardless of the household situation, humans tend to pick up partners that are going to have the same life expectancy that they have.

We all seem to have a preponderance towards a certain type of accelerated aging.
Polina Mamoshina


Lee: I think that’s surprise, I thought you would have picked a partner who has got maximum life expectancy?

Polina: Right. But it’s also maximum to you. I mean, not maximum and you will have a shorter lifespan. No it’s like, closer to the one that you will have. And that’s really interesting.

Lee: But I wonder how that also works when, say, the man who’s 20 years older than the woman?

Polina: Right. Which is known to make men live longer. Yeah.

Lee: Oh. That’s good news. Yeah, because you’re swapping plasma. We’re starting you talking about blood plasma exchanges next. Hey, by the way, that would be a really good intervention to test, is the plasma exchanges with youthful blood.

Polina: Yeah, but it’s a bit tricky to test because here it’s a bit harder to separate the effect on the blood markers from the actual effect on acceleration of aging rates. So probably there, you actually need to go for other markers, not blood based. Something, I don’t know, skin based, muscle tissue based, and so on. Because, again, you’re kind of messing up a lot of things there, changing a lot of levels of blood analytes and [inaudible] concentration in blood.

Lee: I’ve been very lucky in that I’ve had some world leading scientists give me input and they have definitely steered my mind towards focusing on three areas for, quote, best clocks, going forward. And it is transcriptomic, proteomic, but also glycomic, which is why I had GlycanAge on an earlier podcast. But there’s only two labs doing high-through glycomics, as far as I understand.

Polina: Right.

Lee: So that’s an exciting area, also, is glycobiology. Which I was completely unaware of until recently. Have you looked at any glycobiology? I mean, I was completely 100% unaware of it.

Polina: Yeah, I mean, with all the advanced indication products, they increase with age. Their levels increase with age. And for sure, it’s something you should be aware of and it’s something you expect to be reflective of someone’s chronological and also biological age. But then, as you correctly said, there are not that many labs who can analysis, and that’s actually downside of our approach. We have to have a lot of data, plus, we want to be really massive now. And even with transcriptome and protome, it’s tricky. So right now…

Lee: Yeah, I wondered when you were getting your data for that.

Polina: Yeah. We’re going to but we have to find providers who will do the analysis for us because we are, ultimately, dry lab. So we don’t do any wet experiments. And that’s why we’re partnered with clinics, and in the end we aim to extend the panel. And plus, we believe that it will be the way for us to democratize it. So we’re going to have something really massive that will allow us then to have certain types of [inaudible] profile, only limited number of individuals who are really interested in profiling them and doing such a deep dive. But then it will help us to make the technology cheaper so we can bring it to the broader audience.

Lee: It’s going to be an exciting next 10 years.

Polina: For sure.

We don’t believe there is a one universal marker that can be used for completely different tissues, organs and systems in the body.  
Polina Mamoshina


Lee: So I’m super excited by it. I’d better finish off. I feel a touch guilty. Maybe I should feel a lot guilty because I’ve overran the promised time with you. So I’ll finish off just asking about… Do you see the rise of longevity clinics?

Lee: And also, tied to that, slowing aging, reversing aging has been something that’s been promised for, well, I think I would definitely say, the last 15 years there’s definitely been quite a bit of amplitude to those promises. But what I’ve gathered, looking at your materials, is that you actually believe that this is coming to fruition now, something that’s not just a promise or a dream. And I think that would be an area to close out on.

Polina: Right. I totally agree with you. So I believe there is an emerging trend there. There are several clinics out there who actually already using aging clocks in their care. Of course, they have to be really cautious with that but it’s more like a research tool at this point, in terms of how mature the technology is. But they’re still using it. And to do that, I mean, that’s amazing. That’s great.

Polina: And it’s not only great in terms of that now we can actually quantify something, but I believe it also, from a patient perspective, it’s really important outcome. Because you can put patients on interventions but unless they see any effects, they will be disappoint in that and lose interest, and they will never follow up on the treatment and appointments, and so on. And for some of the interventions, it could be a very long time before you will actually feel the effect. And therefore having something that can actually quantify the effect, that could help a lot, I think, from a behavioral point of view, if you have some number…

Lee: Yeah. People’s behavior changes based on a number. Somebody who measures their weight every day tends to lose weight.

Polina: Yeah, exactly. So we believe that you can do the same with aging clocks. So they will be…

Lee: I would agree. One interesting area in the medical side is I highly predict that if people start measuring their biological age accurately and often, those who are on medication, I think they will find many of the medications actually accelerate your biological age.

Polina: Right.

Lee: So that’ll be an interesting interplay in medicine.

Polina: That’s true. That’s true. Some of the cancer drugs, they actually accelerate the cells’ senescence.

Lee: And… Sorry, Polina.

Polina: No, I was just about to say, some of them I XXX phrases I’m using. So, no.

Lee: Okay. And it’s a little bit unclear to me where measurement of biological age ends and health diagnostics begin. The two are quite overlapping, particularly when you look at molecular help, like Q Bio.

Polina: Right. But, again, that’s a tricky question. We have to have doctors aligned with us. And there are a limited number of physicians who actually know the value of biological age. And that will be a long journey before it’ll be in everyday care and before we can actually have it in place. But we hope that we can help to accelerate with those clinics who already are integrating those tools in their care. They will be the one who will actually change the industry.

Lee: I agree. And money’s going to flow into those clinics. You and I can guarantee it. And some decent money, and the money will be increasing. And I think as money flows in, more and more clinics will take on such tools.

Polina: Sure. When they see the popularity, when they see how people will react. Yeah.

Lee: And as we said at the beginning, the COVID situation will accelerate that.

Lee: Polina, sorry for overrunning the time with you. And is there anything else you would like to say or give any URLs to where people can find out more information? Is there anything you’d like to say to close off?

Polina: Right. Thank you, first, for really a great discussion. So it was really great to have this discussion and the great questions. So I enjoyed it a lot.

Polina: In terms of our information, they can always go to our website,, and then we have a lot of materials there. Because, as a company, we try to publish as much as possible, and we have our publications there, some of our explainers, so they can go there and learn more about the app, the Young.AI app. We’re going to launch Young.AI soon.

Lee: So this is beginning of October. Do you have any idea when the Young.AI app will be in the App Store, the app that lets you test your biological age by taking a photo and by entering more bloods and so on if you wish to do that?

Polina: Right. So Apple’s a bit tricky. So we no control if… So it could be any day now. We are in review. So it could be any day now. But we don’t have an exact date, I’m afraid.

Lee: Well, it’s greatly appreciated. Viva la revolución.

Polina: Yes.

Lee: Hey. Thank you so much, Polina.

Polina: Thank you.