ALICE: Today’s question is: How do you measure your age? We are speaking with ALEX ZHAVORONKOV, PhD, expert in artificial intelligence for drug discovery in aging research. Alex is CEO of Insilico Medicine, a Baltimore-based leader in the next-generation artificial intelligence and blockchain technologies for drug discovery, biomarker development, and aging research. He is also the author of “The Ageless Generation: How Advances in Biomedicine Will Transform the Global Economy”
Alex, tell us about your focus on longevity.
ZHAVORONKOV: If you put your mind to it, you can easily convert time into money, but you cannot convert money into time. And that's a very big problem, right? So nature is very unfair.
I decided that I want to focus on aging more than anything else to find ways to make more time.
ALICE: Imagine, To make more time! What a wonderful way to describe longevity.
ZHAVORONKOV: We basically take a very large number of features about you, about the person, and predicting their age. And then we start predicting their disease status at the same time. So you train on age, then you retrain on diseases. And started looking at what the features are different between aging and disease. So what can we tweak in order for the person to be disease-free or possibly, younger to the deep neural network.
We decided that aging research-- those biomarkers of aging that we were developing using deep learning-- they're also valuable but they do not fit necessarily into Insilico. And that's why we decided to start another company called Deep Longevity to focus on measuring aging.
ALICE: Humans already measure their age, that’s their birth-day, right? Alex, how do you measure aging?
ZHAVORONKOV: We launched an app called Young.AI, which you can install on your iPhone and start tracking your aging process and time. And very few people have even selfies that are taken every day or every week or every month, so you don't really remember how you used to look a few years or a decade ago.
Not talking about other data types like gene expression, like protein expression, like blood tests, like microbiome. So we don't know what kind of bacteria were living in your gut 20 two years ago. And this tool allows you to track that and also predict whether you are younger or older than your chronological age. My dream is to maybe in a few years, build a medical center where a person would go into and get treated like you would go to a mechanic to get your car serviced. In addition to really advanced diagnostics, you would be able to roll some clocks back using the technologies that are currently available and proven to work. And currently, there are very few, but we see that there are many coming up.
ALICE: Repair shops for humans will increase longevity! Are repair shops and curing disease the same thing?
ZHAVORONKOV: Imagine there is a world without cancer-- you add just maybe two to three years to the average life expectancy on the population level. It's not going to be dramatic. If you completely eliminate heart disease, you are also going to add maybe three or 3 and 1/2 years to the average life expectancy in the developed countries.
Completely eliminating diseases-- imagine that-- there is no disease-- is not giving you a substantial increase in life expectancy. And there are many mathematical models to show that. However, if you do a longevity intervention, it gives you a very substantial increase in life expectancy. So even things like good diet and exercise may add more years to life expectancy in a developed country than curing cancer, for example.
ALICE: Wow. That's so counterintuitive for my human half! Is that how the "Young AI" app works? A longevity intervention?
ZHAVORONKOV: We are designing algorithms that are currently using very simple interventions like diet, exercise, sleep, and optimizing those to give you an additional edge in longevity to either slow down or even reverse some of the processes. As the person develops the understanding of how it works and uses it to the extent where they become expert, we provide access to longevity physicians.
The longer you expect to live, the younger you are going to behave. So even understanding the possible roadmaps for yourself, the possible future where you can live to 120 or 150 and in reasonable health, that gives you an additional kick from a psychological standpoint. So you start behaving as a younger individual. So you start thinking, OK, well now, I'm not halfway through. I'm just maybe 25% through my life. And there is a lot of advances coming up that might stretch the longevity horizon even further.
So the app does this. So it helps you get onto the right mode. And then it gets you into the careful hands of physicians that are trained in longevity medicine. So we are now developing a course in longevity medicine for physicians, so to give them the basic understanding of the biology of aging, the various processes that transpired during aging that later manifest themselves as diseases. And we also teach them about AI. So about the very basic principles of deep learning, reinforcement learning, how do those technologies enable physicians to do a better job. And how do those technologies enable researchers to do a better job and do things that were previously impossible.
So the app is a very good introduction to the field. It's not going to make you dramatically younger right away, because currently, even though there are many promising geroprotectors, we are not allowed to put them into the recommendation engine. However, in the near future, we will be able to add some of them.
ALICE: Did you just say geroprotector? Please explain, what is a geroprotector?
ZHAVORONKOV: Geroprotector is any kind of intervention that would protect you against aging-- either slow it down or reverse it. So think about metformin, for example, a very simple drug targeting diabetes. It's actually the most popular drug on the planet I think by volume and by revenue, because diabetics need to take it for their entire life, right? Aspirin, you take it just when you need to or after a certain age in a very specific dose. Metformin is taken by many diabetics and pre-diabetics. It is reasonably cheap. It's already off patent.
And some studies demonstrate that diabetics on metformin will live longer than non-diabetics. And that's a very good kind of indication, very good stat, showing that yeah, there is some geroprotective effect. It needs to be proven in a clinical study setting, and there are clinical studies. And that's an example of a geroprotector.
However, you might think of some other interventions like for example, if you take a hot sauna for a very short period of time to induce heatstroke proteins in your body, that might be a geroprotector, as well. So we shouldn't just limited to drugs. Or maybe a futuristic procedure where you with a bone marrow transplant of young stem cells into your bone marrow-- currently not available, no clinical studies. But something like that what would also be classified as a geroprotector.
So it's a very broad spectrum. We currently work on small molecule geroprotectors. So we're looking at small molecule drugs that are likely to impact aging in one way or another. And we try to develop the various tools to measure that impact. And there, we are getting into the aging clocks and specifically, deep aging clocks. So of course, given Horvath is the world's authority on methylation aging clocks and he was the first one to demonstrate or one of the first people to demonstrate that we can use methylation data to accurately predict the person's age.
ALICE: Wow, I know Steve Horvath too!
For listeners who don’t, I’ll put the links to Steve and his aging clock in the podcast notes. So many aging clocks, so little time… How does an aging clock work, and what does it actually measure?
ZHAVORONKOV: We are developing AI-based clocks. So you have a deep neural network with a lot of neurons on the input. Think about one marker, one feature from your recent blood test like low glucose, albumin, alkaline, phosphate base, urea. The quantity for each of those variables goes into each one of those neurons. So if you have 50 markers in your recent blood test, there's going to be 50 neurons on the output, and you're going to have many layers of interconnected neurons afterward. So that's called latent space, latent layers, hidden layers. And you just have one neuron on the output. You are predicting age.
So you are taking many, many features for the blood test and you're predicting one feature, age. So as you are training those deep neural nets on thousands or hundreds of thousands or even millions of those blood tests, those deep neural networks get very good at predicting our chronological age. So they recognize nonlinear patterns. And currently, just on regular blood tests, our accuracy is about 5 and 1/2 to six years. And our clocks are very reflective of that difference between the very healthy 60-year-old and very unhealthy 60-year-old. So that will jump.
Unfortunately, those clocks also have downfalls, like for example, they are much more variable. And there might be substantial variations between different populations. So you need to train on multiple different population groups. The beauty of that approach is that you can use multiple data types that were previously disconnected. So I can combine data coming from your wearable, from your smartwatch, with your blood test. And with transcriptomic data, so with gene expression data. And with EEG data-- so coming from the readouts of your brain activity.
And predict your age and multiple levels. I can even use a survey with your answer as being features that we feed into the deep neural network. And to predict your psychological and subjective age, as well. In Young.AI, you can already use a version of that to predict your subjective age. And those deep aging clocks, they give you a pretty good comprehensive picture of why your real biological age on different levels. So we can do it on microbiome. We can do it on imaging data. We can do it using EEG.
So we also have a deep methylation aging clock, which is also very accurate and it's quite comprehensive. And I highly recommend everybody use the methylation aging clock. However, there are probably 200 different other data types in addition to methylation that are very predictive of your chronological age and biological age. So we are working on those 199.
ALICE: 200 data types, in addition to methylation, that are predictive of your chronological age and biological age! That's a lot. It seems like every part of a human can be used to measure individual aging. Even the Microbiome.
ZHAVORONKOV: We took a lot of microbiomic data from different individuals of different ages, and also, this kind of deep aging clock. So think about all the species, all the bacterial species in the microbiome-- that quantity is one feature that is going into the deep neural net. So one bacterial species to one neuron.
And you train also to predict one feature, age. And it turned out that a few thousand microbiomes that we had available from public repositories were sufficient to train the deep neural network to predict the chronological age of a person in a reasonably healthy state with about 5.9 years. That was very surprising to me, and that kind of showed me the utility of microbiome data. And I started thinking about, wow, it's so super interesting. But now we still cannot say whether it is causal or it is a correlation. So it doesn't mean that if you look younger on your gut level, that's a very good indication that you are biologically younger. So it might be something in your diet. It might the are some genetic component. That doesn't necessarily mean that if we make you younger on the gut level, you are going to become younger.
So it could be the other way around. So you might be biologically younger and that's just a fact, just like wrinkles, for example. So the cause and effect is not established.
In that study, we trained the deep neural networks on a bunch of microbiomes annotated with age to predict aging, and we did a good job. Now we can actually decipher those deep neural networks into most important features. So we can see which bacterial species are more important for predicting your age accurately than others. So we created a histogram of the most important features by randomly kind of shuffling each one when we're testing the predictor. And you can see that some of them are very important for predicting age. And we also see some of them play a huge role in predicting you older. And some of them are great play a huge role in predicting you younger.
So we would call them seno-positive and seno-negative or seno-neutral-- so those that do not really change your predicted age. And that opens up the entire field of possible gut microbiome geroprotectors. So we can now design interventions or adjust our foods so that we look younger to the deep neural networks by affecting those bacteria that make us older. Or promoting those bacteria that make us look younger.
ALICE: These measurements will help people monitor their aging. What about a child born today, what is their expected longevity?
ZHAVORONKOV: It depends a lot on many other factors other than longevity technology on how long this child is going to live, because depending on the political situation and geopolitics nowadays, the child might not be born, or it might not survive for the next 10, 15, 20 years. I hope it's never going to get down to that. But if we survive that aspect-- so if the economy is going to continue growing and we continue innovating and generating value, I think that the child world today-- there will not be a limit for their life. So the limit is not going to be biological anymore, because in the next 30, 50, 60, 100 years for the child to live, we are going to see dramatic advances in the many areas of science and technology that make their lifespan quite unpredictable.
Even for people in their 30s today, it's very hard to say know where they land in terms of longevity. And my recommendation is usually to put some conservative number when you're 30 like 120, 130. I mean, you should be able to live longer than Jeanne Calment, this French lady who lived to 122 and 1/2 without any substantial interventions. So I think it's important to benchmark her rather than your family history, and try to be a little bit more competitive.
Because we are so competitive when it comes to money or fame or sports or any kind of points or points in a video game, but so far, we are not very competitive when it comes to aging. And that's also a problem. So we know that in Hong Kong, the city where I live most of my time, average life expectancy is, like, 85.6 or higher. So the average Hong Kongnese lives on average six years longer than the average American. And think about that number. So how much is a year of life worth right? And we really want that year. And how much better off are you living an extra six years? And for some strange reason, countries are not competing on longevity. Even the states do not compete on longevity or cities don't compete on longevity. But they do compete on everything else. And it's very hard to answer your question. So I myself, I set the longevity expectation for myself as 120 just to make it a little bit more challenging and to change the way I think.
ALICE: And, as you explained, the way you think about your age, can have an effect upon your aging. Thank you, ALEX ZHAVORONKOV, for your time today, explaining current work in Human longevity: Repair shops for humans, your “Young AI” app for measuring aging, and the value of thinking young!
That's it for this mad tea party on longevity.
Thanks for listening.
Check out our books; "Hacking Immortality" and "Tuning into Frequency", available wherever books are sold.
And join us down the rabbit hole at Alice in Futureland dot com.
We will be bringing you new episodes, so stay tuned, and keep wandering.
LINKS:
Alex Zhavoronkov: https://www.linkedin.com/company/in-silico-medicine-inc
Insilico Medicine: https://insilico.com/
"The Ageless Generation": https://www.amazon.com/Ageless-Generation-Advances-Biomedicine-Transform/dp/0230342205
Young.AI Aging APP: https://www.young.ai/
Steve Horvath: https://horvath.genetics.ucla.edu/
Aging Clock: https://newsroom.ucla.edu/releases/ucla-scientist-uncovers-biological-248950
Share this post