Generic selectors
Exact matches only
Search in title
Search in content
Post Type Selectors
Generic selectors
Exact matches only
Search in title
Search in content
Post Type Selectors

MOSAIC: Multi-Organ Scores of Aging across Integrated Components

Listen or watch on your favorite platforms

Various aging clocks have been developed to quantify the aging process and predict age-related diseases. These biological age clocks are powered by different types of omics data and clinical biomarkers, and they’re especially useful for observational studies, clinical trials, and basic science aimed at combating biological aging.

Nonetheless, current research indicates that there is significant variation in aging, with deterioration and diseases affecting different organ systems and functional domains at different rates among individuals.
While existing aging clocks can measure variations in the degree of aging, they do not account for variations in the way that aging occurs, such as in specific organ systems or functional domains.

This is exactly what Raghav Sehgal has been working on during his career at Yale University – biological age clocks for 11 organ systems such as immune function, metabolic function, hepatic function, cardiac function, renal function and more.

Knowing the age of your organs can provide several advantages over knowing just your biological age. Some of these include:

1. Better understanding of disease risk: Knowing the age of individual organs can help identify which organs are aging faster and therefore at higher risk for developing age-related diseases. This information can be used to develop targeted interventions to prevent or delay the onset of these diseases.

2. Precision medicine: Understanding the age of specific organs can help tailor medical treatments to an individual’s needs. This can improve treatment outcomes and minimize side effects.

3. Earlier detection of disease: Changes in the age of specific organs may be an early sign of disease. By monitoring the age of individual organs, it may be possible to detect disease at an earlier stage when it is more treatable.

4. Improved health and lifestyle choices: Understanding the age of specific organs can help individuals make better lifestyle choices to improve organ health. For example, if a person’s liver is aging faster than their chronological age, they may be motivated to adopt a healthier diet and lifestyle to improve liver function.

In this week’s Everything Epigenetics podcast, Raghav and I chat about his novel epigenetic aging clock called the “Systems Aging Clock” which is based on a combination of epigenetic changes and organ and bodily function-based mortality indices. We focus on his research which has shown that the Systems Aging Clock is a more accurate predictor of biological age than other existing epigenetic clocks. We also discuss how his system aging scores and aging clock is validated in different clinical studies to show the added advantage of such a measure, such as the fact that people with similar epigenetic age may have very different system scores. Raghav’s work on the Systems Aging Clock has the potential to advance our understanding of aging and to provide new tools for predicting and managing age-related diseases.

Raghav is a PhD student at Yale University presently solving Aging using deep learning on multi-omic and multi-granular data.

In this podcast you’ll learn about:

– Raghav’s background and how he became interested in biological systems and aging
– The idea that a single number (epigenetic biological age) might not provide enough feedback for the heterogeneity with aging
– How each of our organs age at different rates
– Ageotypes (aging subtypes)
– Using the age of your organs to guide preventative treatment
– Raghav’s systems based approach to aging using “different organ systems or functional domains”
– The limitations of the epigenetic methylation data currently available
– The reason behind choosing the 11 organ systems
– How Raghav models unique epigenetic aging trajectories from 11 distinct groups of organ systems
– Organ systems Raghav would have liked to include in this research but couldn’t because of limited data
– The reason behind making an aging clock which combines the 11 organ system scores
– The value of the 11 organ biological age clocks
– Issues with particular organ system scores
– How Raghav accounted for the issues
– How he validated his system aging scores and aging clock
– Applications of the systems aging clock
– Why you should look at your systems aging clock over traditional Epigenetic Age clocks


Raghav Sehgal

hannah_went (00:01.196)

welcome to the everything epigenetics podcast rug of i am so excited to have you today we have a lot of exciting things to chat about


raghav_sehgal (00:11.13)

Oh, thank you for having me here, Hannah. This is really exciting. And yeah, I’m very excited to chat with you today and, I guess, answer all your <br> questions.


hannah_went (00:21.416)

yeah i was just so curious about your work because a lot of it isn’t available to the public i’m very new myself to all of your research and super keen to learn more i spoke with several people after they attended the gordon research conference as i mentioned to you when we were just chatting and they talked very highly about the work that you’re you’re doing and before we dive head first jumping i’d love to start by just getting to know you a little bit more so tell me a little bit about where you you are today


how you got there you your interest and in where that really peaked


raghav_sehgal (00:55.49)

Yeah, so I’ve had a rather interesting journey and sort of tumbled my way into sort of the aging field in all honesty. So I did my engineering back in India. It was in electronics and communication engineering, a very different field. And during my undergrad, actually, I came across this really cool company called Elucidator. They were just kind of starting out and they needed someone


data scientists. They were a computational biology company, sort of agglomerating publicly available data sets and selling them to pharma companies. And that’s kind of where I sort of started getting involved in computational biology, vaguely speaking. And so I spent a couple of years of my life sort of working with them. The startup now is a fairly big startup. It’s a series they funded $100 million startups. I’m fairly proud of it. And


During that period, I also realized though that I have sort of, I’m mathematically very strong in sort of doing something like computational biology, given my engineering background. But when it comes to the actual biology of things, that’s not really my strong point. And which is why I thought that it might be a great opportunity for me to sort of take this up in grad school and sort of invest more time into it. And so that’s what I did.


biologist. And initially my plan was, you know, I’ll be working with like big omic data sets and finding new drugs and new drug targets. But on a very interesting seminar, which I wasn’t planning to go to at all, a friend of mine dragged me to it. And just like any grad student, since there was free food, I agreed. This seminar was actually being led by Morgan Levine, who was


hannah_went (02:48.296)



raghav_sehgal (02:55.45)

then a PI, a professor at Yale. And yeah, she just kind of changed my perspective on aging. I never saw aging as like a disease, so to speak, or I always thought it’s kind of this natural process that happens to everyone. And I think that perspective change really drew me into the field and kind of ended up with me kind of rotating with Morgan and eventually joining her lab and sort of working on building all of these really cool


working on.


hannah_went (03:27.416)

yeah that’s awesome that you saw her speak and that was like your turning point and the career and then you joined her lap then it opened you up into this entire new world because i remember that turning point for me as well where i had no idea this this existed and when people ask me about that i can remember when i was opened to this world of aging and preventative integrative functional medicine so it’s like weird to think of my life before when i didn’t know about it it’s like all these possibilities


raghav_sehgal (03:54.67)

Yeah, it’s really transformative in that sense, right? Like, I always feel like some of these, you know, in science, you always kind of feel like, hey, I’m doing this really cool thing, and this is kind of the best thing that the world can see. And then you experience this new kind of science, and you’re like, oh my god, I didn’t know this at all. And I can totally relate with that experience of yours, where you’re like, mind blown away.


hannah_went (03:57.396)



hannah_went (04:20.036)

yeah such an epiphany well no i love hearing hearing about your back story always curious to see how people really get get started in this field so we’re going to talk about today is more of your systems aging clock and i want to preface by saying a lot of our listeners are probably very familiar with a lot of my listeners are probably very familiar with these epi genetic clocks you know these biological aging clocks and those are based on large scale meta analysis they come up with this


biological age that the single metric sometimes you know there’s the pace of aging that denindinclock but you know your work is influenced by this idea that the single biological age number is just not enough and may not provide enough feed back to what’s happening on a biological level so you know you believe that this has to do a lot with a heterogeneity of aging and i want to really get into that you know have you talk about


that may be defined heterogeneity to begin and then we’ll go from there


raghav_sehgal (05:24.19)

Yeah, no, that makes sense. So I’ll kind of start with kind of the clocks that are out there. I think they’re amazing. You know, Grimmage, Dunedin Pace, Spinoa Age, also like the first generation of clocks that were kind of predicting age and not sort of something like mortality or pace of aging. I think they’re all amazing clocks and they have provided us with such meaningful insight into how we’re aging as individuals and sort of their relationships with different sort of these age-accelerated relationships


hannah_went (05:30.476)



raghav_sehgal (05:54.41)

comorbidities has been quite impressive and prior to sort of these clocks there was never anything that could quite predict aging like these clocks can. So to start with I stand on the shoulders of stalwarts in a way right these clocks are amazing and they are so powerful that being said you know there’s always space for improvement in anything that’s out there and I


we really started thinking about this. So these clocks, if conceptually speaking, what do they do? They take sort of whole blood methylation, right? And they use it as a proxy of your whole body aging, which is quite impressive. These numbers that we get are quite impressive. But here’s the thing, as we know from biology in general, our organs and our different biological systems don’t age at the same rate.


And if all of our biological systems were aging at the same rate, all of us will get the same diseases at the same time. So all of us will get cardiovascular disease at 60, and then all of us will get Alzheimer’s at 80. But that’s not the case. Our genetics, our environmental factors influence the kind of aging that occurs in our body. And that leads to this differential aging across our different organs and biological systems.


heterogeneity, which is what I’m talking about. So at one level there’s heterogeneity, which is across different individuals, aging rates are different, but within an individual itself, the organs and the biological systems age at very different rates. And so what we really wanted to capture was how are people or how are the different organs within an individual aging differently?


raghav_sehgal (07:54.15)

Because now using this information, we believe that we can actually more accurately predict the kind of comorbidities, the kind of health span, the kind of lifespan a person would have. So hypothetically speaking, think about two individuals who have the same overall biological age, say calculated by Dunedin pace or Grimm age. But we’d really like to understand why do they have the same biological age, right?


because both of their hearts are really aged. Maybe one of their hearts is really aged, for someone else, their liver is really aged. And then the kind of interventions that you practice in this case would be very different. Say, if I have a really strong heart aging or really accelerated heart aging, I might wanna take metformin, right? Hypothetically speaking, assuming if metformin does it, decrease cardiovascular age. Or if my brain aging is really high, I might wanna take nicotamide,


another one which is can slow brain aging. But I’m not saying you should take metformin or nicotamide for any of those reasons. But at the same time, I am saying that, you know, you can then practice more targeted interventions that can potentially lead down to slowing in specific organ systems. And that hopefully will be a lot more meaningful in actually being able to sort of decrease your overall biological age.


hannah_went (09:22.556)

you very interesting so was that a new thought that all of these organ systems are aging differently or was that something that’s been around for a while and that we were maybe a little bit aware of but never have done much research around that


raghav_sehgal (09:27.95)

Thanks for watching!


raghav_sehgal (09:36.11)

I think we always kind of knew at some level that our different organ systems will age differently. You know, there’s canonical evidence in the fact that, as I said, I was giving the example, not all of us die of the same disease at the same time, right? So clearly our organ systems are aging in different manners. And so there’s always been kind of this canonical evidence, but for once, I think we have been able to quantitatively show that that’s the case,


to age differently in their different organs and those different rates of aging are actually sort of quantifiable in a way and from a very simple method relatively speaking.


hannah_went (10:18.496)

yeah i’ve measured my methalation market s on the epic fifty several times i’m waiting for my most recent data to get back but now i want to know the age of my organs like that’s the next best thing because sure if you’re aging if you’re biological ages is older than you would definitely want to know what organs are maybe causing that acceleration but if you’re biological age is younger my assumption is you could still have some organs that are maybe aging a little bit older as well so you would want to identify those organs that may be having


a little bit of a quicker aging and take set interventions or preventative approach to then no further decrease that biological age


raghav_sehgal (10:52.013)



raghav_sehgal (10:55.31)

Exactly, exactly, right? And I think what’s really valuable out of this is that, and then probably we can talk a little more about this as we move forward, but beyond just the fact that, you know, different organs age differently, I think there’s also like a combinatorial effect. So think about it like when a car starts to break down, if your axle goes off or if your pedal goes off, right? Or if your brake goes off,


or starts bearing off, that’s not the only thing that will wear off. It will have like a domino effect of sorts, right? And so you will not only just see that, hey, your heart aging is going really fast, but maybe your kidney aging, which is very closely tied into heart aging, is also accelerating at the same time. And that actually gets into a really interesting sort of caveat or of aging that we are now starting to call


hannah_went (11:33.796)



raghav_sehgal (11:55.25)

subtypes or ageo types almost, right? That individuals, if you actually start looking at these sort of different measures that we have created and look at them together instead of independently, you can actually start seeing individuals being aging in almost very different patterns, having very different predisposition to diseases. And that’s where it really gets interesting because that is when you can really start intervening and saying that, hey, this


these medications so that we can slow down your aging subtype. And that to me is like really valuable.


hannah_went (12:32.716)

so now you may have these aging organs which is going to affect cenergistically all together these organs right think of it all out on like a string and they’re pulling each other back and forth and then based on that you’re classified or put into some type of a go type which could eventually have a treatment plan and it all goes back to that preventative type of medical approach you know i really like that and let’s get more on the computational side of what with your work since


raghav_sehgal (12:38.891)

Good night.


raghav_sehgal (12:48.557)



raghav_sehgal (12:53.831)

Exactly. Yep.


raghav_sehgal (13:01.087)



hannah_went (13:02.416)

really your your bread and butter so we have these different organ systems or these functional domains that you call it and you’re using this system based approach that we’ve been chatting about so can you explain like how you worked through through the body like what organ systems did you pick you know what made you choose those specific biological processes


raghav_sehgal (13:20.05)

Yeah, so from the get-go, what we did was we started with a more unsupervised approach where we said that, hey, what we’re going to do is we’re going to take a lot of these functional biomarkers and clinical chemistry biomarkers and cell sorting information from different individuals, and we are all going to throw it into like a dimensionality reduction technique and see what dimensions show up.


raghav_sehgal (13:50.05)

into that these different biomarkers were kind of organizing themselves or sort of having similar dimensions to each other, depending on their on the sort of organs that they were related to or the biological process they were related to. So in practice, this was kind of almost like, you know, it was kind of like its own discovery that they were organizing themselves into these sort of behavioral patterns that were marked by these organs. And so that


that, hey, there are these different systems that we can really sort of are behaving very distinctly in a way, and we can really capture that information separately, at least based on very well-known biomarkers. And then the question was that, hey, if these systems are behaving very differently and have these sort of very distinct sort of trajectories to speak of, then is this information then being


raghav_sehgal (14:50.29)

And lo and behold, we built these kind of elastic neck models to sort of predict these sort of trajectories of these organs. And they do actually segregate out. They are very different as they are not sort of highly correlated and behave very differently. And actually, then we sort of even validate to show that they are very specific to the organ that they are built for.


It was, partly it was like, initially it was just like, we tried things out and we saw that there were these 11 organ systems, biological systems that were showing up. And that being said, I also do sort of recognize that this is not the end all be all, that there are only 11 systems. It’s partly a limitation of the data that we have as well. We have only so many biomarkers, even though we use like 160 biomarkers or something,


biomarkers only. If we had hypothetically even more biomarkers or even more data, we might probably discover even more biological systems that were segregating out. But given the data that we have, we were able to find sort of these 11 distinct systems and which we then trained using our methylation data to figure out these aging tragic trees.


hannah_went (16:12.096)

sure yeah i imagine it’s very hard when you’re looking at something so complex such as the human body and trying to break it down into these different systems or classify it right like whenever anyone asks hey what do you do it’s just so hard to talk about like biological age union do you name methalation and like where to start so i couldn’t imagine like trying to break down the body in these different ways and just to name a few correct me if i’m wrong i think you have some like the immune function metabolic function hepatic function


cardiac reno and obviously you know a couple more you said eleven hundred sixty pile markers that fall into each of those those categories um were there any systems you would have liked to include that didn’t make the list that you were just like oh darn and i know you said you know you’re working with data that you have there of course which is um what all of us need to do when when creating a new project or looking at a new area of interest


raghav_sehgal (17:08.39)

Yeah, no, I think in that sense, systems age is fairly sort of very well encompassing. That being said, you know, there’s always, as I said, scope for improvement in science. So as a limitation, I can tell you is that even though we have, for example, like a hormone system, we don’t necessarily have a system that is able to capture sort of pre-productive aging, which I


raghav_sehgal (17:38.25)

especially in women and sort of in people of the female sex. And so I think it’s one other sort of system that I would really like to capture in the future because that would additionally sort of improve our understanding of the biology of aging in general. Another system that I feel that we could potentially additionally capture is kind of, again, the equivalent of that


system. But another one which I feel is very important that could be additionally captured. So we have like a metabolic system and we have like the liver but we don’t have something which is very specific to the gut. And I know that recently if you have seen the hallmarks of aging paper, dysbiosis has become like one of this new hallmark of aging. Sort of the gut microbiome has become this new fulcrum point of aging. And so that is one other sort of


system that I would really like to capture at some point using this data. But again, there could be so many more that are out there and as we have more data we discover more of them. But yeah, I think just from the get-go these are a few that I really think would be exceptionally well added to this list of system scores that we have.


hannah_went (19:01.916)

sure yeah i know you know fertility obviously is a huge studied kind of sub group or category within gnmetholation and then it was interesting and funny that you brought up the microbiom in the gut because i asked that question a lot for my health care provider is like you know the relationship and metholation and there’s obviously definitely something there but it’s not really well studied with methalation so i think a lot that we can uncover in and learn so very cool


raghav_sehgal (19:25.416)



hannah_went (19:31.816)

have these distinct groups of a biological age processes so you’re taking these eleven systems and you so you have these biological scores right for each of these systems so you kind of have do you have these biological age clock then for those eleven eleven systems correct


raghav_sehgal (19:47.831)

Yeah, that is correct. Yeah.


hannah_went (19:49.816)

perfect and then you’re taking those and you’re combining them to create the system’s aging clock right


raghav_sehgal (19:55.45)

Yeah, that is correct.


hannah_went (19:57.696)

perfect and what what did you find there a yeah explain that process


raghav_sehgal (20:00.47)

Yeah, no, I think so. So partly, I’ll be very honest, I think the reason why we completely made sort of this overall aging clock, I think it’s really useful because we would like to sort of compare it to the presently available clocks that are out there and show kind of whether we are able to perform as good or not and so on and so forth. But I think it


raghav_sehgal (20:30.15)

individual scores that are really, or these individual sort of organ clocks that are the really meat of this whole exercise that we have done, the scientific approach that we have taken. And so the overall clock is really helpful in sort of telling you, hey, this is your overall age, this is what you look like, but then it is these kinds of independent scores that really inform you that, hey, this is why your overall aging score is so high. So if you get, say,


years older than you are or supposed to be. The reason you’re two years older is because your heart is four years older than it’s supposed to be, but your liver is two years younger. So maybe don’t focus on your liver, focus more on your heart aging and so on and so forth. And so that’s kind of really where it gets super interesting for me. I think the other thing that we really found and I think that’s very valuable is, so a


raghav_sehgal (21:30.15)

available clocks that are out there, and they’re really good. Like phenol age, grim age, Dunedin pace, they can capture, so the way they are built, they capture sort of these different aging phenol dimensions really well. But that’s kind of the thing. They’re able to capture maybe one or a few aging dimensions really well, but what we wanted to attempt was build a clock that is able to capture all of these different aging dimensions really well, and kind of have a more


balanced approach of sorts, which is not to say that any of those clocks are limited in any way. I think they’re all exceptional clocks. You know, Grimage predicts for cardiovascular disease and lung cancer really well. JuniDenPace predicts for sort of musculoskeletal related issues, sort of frailty really well. PhenoAge predicts for a bunch of other things really well. But what you really want is kind of like a balanced core across them all. You


over predicting for one particular dimension. What you really wanna be saying that, hey, my clock is not really over predicting for cardiovascular aging dimension excessively, but it’s also predicting well for kidney dimension of aging or some other dimension of aging. And that’s what, in some of our analysis, that’s what we have really seen that what systems age because it’s kind of using all of the systems approach of like integrating all of these different informations,


capture all of these different dimensions of aging in a more balanced fashion without being kind of biased in one particular direction.


hannah_went (23:09.876)

and i think the


i agree with you all of the clocks that have been created today the pgnetic clock like we wouldn’t be where we are today if we didn’t start out with that basic corvet three hurd fifty three c p g first generation clock right and that’s very useful for certain circumstances depending on the outcome right so i think the limitation with the research in this field is when you’re only looking at you know one clock right we want we need to to really understand what agin is an you know the causation rather than correlation i think we need to look all of the


raghav_sehgal (23:22.914)



raghav_sehgal (23:27.414)



raghav_sehgal (23:35.271)



raghav_sehgal (23:41.514)



hannah_went (23:43.176)

clocks across the board so that’s why yours is even more unique because you have this you know systems aging clock but then you have these like i guess you could call them sub categories eleven clock stemming of


raghav_sehgal (23:52.29)

Yeah, yeah, I almost call them like different aging dimensions, if you think about it, right? There’s different aging dimensions of your body, because you’re not aging in one dimension, you’re not sort of unidimensional, so to speak, often your biology. And that’s what’s really interesting. And I guess that’s what’s been sort of previously been missing, right? Like all of these clocks are amazing, but they were capturing a handful of dimensions really well.


hannah_went (23:58.376)

i like that


hannah_went (24:22.916)



raghav_sehgal (24:23.072)

Now, since they’re all separated out in the systems-age framework, they are right in your face, I guess.


hannah_went (24:30.036)

yeah definitely and one question i had when you were talking about how you created these clocks what what cohort of people did you use for for your study or what did that look like


raghav_sehgal (24:39.41)

Yeah, so we had two particular training datasets that we used in two different steps of our training process. So we had the health and retirement study, which had about 10,000 plus individuals that we used for sort of categorizing our different biomarkers and then sort of building these predictors using the epigenetic data. And then in the Framingham HOT study, which had about 4,000 something


individuals or a little less than 4,000, we used it to sort of train these scores and the clock ages and eventually the whole systems age as well. So we use both of these for our training purposes. And then we did a lot of validation as well. So we went to sort of the Women’s Health Initiative study, where we sort of showed a lot of our insights around and again, Women’s Health Initiative was kind of like a blind data set for us.


showed the improvements that these clocks really provide and kind of the agiotypes information. And that had about 5,000 samples. And then we’re also sort of looking at the SATSA data set, which is a more longitudinal data set to show that longitudinally how these things change.


hannah_went (26:00.016)

yeah and i don’t even know if i’ve spoken with anyone about this yet but can you just give our audience a general explanation of why it’s so important to use a different data set for i should say training data set right in a validation data set do you just want to explain that a little bit


raghav_sehgal (26:15.29)

Yeah, sure. I think this goes back to some of the very basic principles of machine learning that you can essentially overfit to a data set, and you can overpredict. And your results might be exceptional in that data set, but they might be completely irrelevant in another data set because your features were selected or your essentially what your methylation in this case


in this particular data but are not really relevant for another data. So what you really want is the data, what you really want to show is that hey, you’re not performing just well in the data that you trained in, but you’re actually performing exceptionally well in a data that you had not touched at all well in your training. And that’s kind of the standard in machine learning, so to speak, which is why we use two training data sets


raghav_sehgal (27:15.25)

speak off in our data while also keeping like a completely two different blind data sets that we have not touched at all during our training data, training exercise, because you don’t want to sort of bias yourself in any way.


hannah_went (27:31.016)

thanks for that explanation just wanted to point that out in case ah people had had questions about you know why would you use a different data set for for training and invalidation i’m going to go back a little bit and ask a question regarding the maybe this was before you decided on the eleven kind of dimensions of aging like you mentioned but did you have a hard time or any issues with some particular biological system scores you know i think you know some metrics may have been imposed in


raghav_sehgal (27:42.872)



hannah_went (28:00.816)

cohort like better brain age prediction may have led to lower heart rate prediction or you know those strings were pulling in opposite directions and ways you didn’t think they would would act and then how do you account for for those those changes or those issues


raghav_sehgal (28:12.73)

Yeah, so I think that’s partly like in some ways the limitation of what we work with is the data that we have, right? And so.


hannah_went (28:19.716)



raghav_sehgal (28:26.53)

Within the limitation of the data that we have, we’d like to assume again, the health and retirement study is supposed to be sort of a cohort that is very representative of the US population. So we would like to believe that we were capturing all potential interactions that could have been there, but that being said, that is not gonna be, it’s only representative. It is not exactly the same human sort of US population.


I, for one, cannot assert and say this with certainty that, hey, we’ve captured all dimensions that this is the best representation or the best ways that are there. That being said, what I can say with a lot of certainty is that given the data that we had and given the training models that we built in our sort of more blind data set, we were able to sort of create the same validations and get the same really high accuracy.


pointed out, you know, there could be issues around, hey, in this particular data set that maybe there’s not enough people who have brain aging or this particular data set, maybe a lot of people have cardiovascular aging. Those are all very reasonable, I think, questions. And I think the best way to sort of solve them is sort of keep improving these clocks with even


raghav_sehgal (29:56.47)

The more people try these epigenetic locks, the better we will be able to build them, the more information we’ll be able to capture. And so it can only sort of improve from now on. And in a lot of ways, right, for us, system age is kind of the beginning of a framework. It’s a framework for people to start really like going crazy on this, right? We want people to, and in one of the reasons


hannah_went (30:21.156)



raghav_sehgal (30:26.81)

to sort of get this out there is because we really want to standardize the pipeline so that anyone can just go and create a clock in whichever dimension of aging that they want to and sort of build atop that as much as they want.


hannah_went (30:43.016)

so i’m going to pick out a piece there that was an amazing answer but you said you just said at the end there normalizing kind of the effects or i just just lost my train of thought there the normalization pathways in using these algorithms that way people can use them in different studies can you explain what that is and what that means because you know something within my company at true diagnostic is what we’ve worked on time


raghav_sehgal (30:56.671)



hannah_went (31:12.896)

time and time again you know mis normalizing normalizing results due to batch effects or due to different issues right there’s a lot of computational biology know science behind that so can you talk about that that’s very new to our listen


raghav_sehgal (31:25.63)

Yeah, yeah, no, I think standardization is kind of the name of the game here, right? Because what you really want is, again, going back to the whole point of using different data sets, going across these different studies is primarily because what you want to build is something that works across different sort of populations, different groups, different ethnicities,


of different socioeconomic backgrounds, so on and so forth. And so unless you’re really sort of standardizing the process through which you’re doing sort of all of these algorithms and building these algorithms, what might end up happening is you’re just kind of overfitting to a data set and you might get like exceptionally good results, but those results are all farce because they have no value in the real world. And at the end of the day, all the science that we do is


patients is to benefit people and if it’s not really, you know, it’s all bogus if it’s not really benefiting them. And so what you really want to do is kind of work or build these algorithms such that A, that they are very explainable, that people can, you know, someone else wants to understand this algorithm and can build a profit. B, they are very reproducible, right? That someone else can really sort of just take the whole thing and retrain it


and again, right? And see what you really want is it should be something that it can be built upon, right? So now that, you know, we have spent these years building this particular thing, I don’t want anyone else to spend another three years just rebuilding this, right? You just take what I’ve done and go crazy with it. Build something even better. I would love that. And that’s kind of the thing, right? You want to standardize the process of these algorithms that


hannah_went (32:58.136)



raghav_sehgal (33:25.63)

So that at the end of the day, all of us want to do good science and sort of build better for the world. And it could be as collaborative as possible if there is a standardization there.


hannah_went (33:37.576)

yes that was the word i was looking for a standardization when i when i ave thought that was broken down perfectly points a b and c you’re you’re really spending so much time on making these usable right for for everyone so i can’t wait to get my hands on them and see them when when you’re done we appreciate all the work that you’ve you’ve been doing for this field so now i went to talk about the applications of of this clock and get into that a little bit


raghav_sehgal (33:40.251)



raghav_sehgal (33:55.313)



hannah_went (34:07.376)

you know i think these clocks would be very interesting in interventional studies obviously right kind of like we hinted at at the beginning and you said you know these application one of the applications of these system clocks and why they’re so important is because you know people with similar epi genetic ages for first generation second generation or the din didn’t pace could actually have different systems aging so what could people use this for like what do you see it being most used for in the future


raghav_sehgal (34:34.61)

Yeah, so I think I see this at multiple different levels. I think at the very trivial and basic level is where it could be something that is very consumer-focused, where individuals go to a doctor and say that, hey, I want to know what my biological age is and what different organ systems are aging for me. And the doctor gives them this single blood jab that gives them a bile of blood and tells them


hey, these are the different organ scores that you have. These are the different rates at which you are aging. And hopefully now you can practice these behavioral or these drug-based interventions that can slow it down, that particular aging. So that’s kind of like the most simplest version of how this can work. The version above that, which I feel is kind of in drug discovery and in sort of building better interventions, right?


way that, let’s take an example of metformin, for example. We know that metformin, at least in retrospective studies, have shown to decrease cardiovascular disease events, right? But do we know for a fact, would that necessarily decrease the whole body aging at the same rate? We don’t know, but maybe it just focuses its effects are primarily on heart aging. So if you run a clinical trial that is capturing


might not see any significant difference. And your whole sort of billion dollar clinical trial goes down the drain. But in reality, maybe there is actually a very specific heart aging that is happening, or a decrease in heart aging that’s happening that you really want to capture. And that is a significant difference. And so what we can really do is kind of build more targeted interventions instead of these more general interventions. And again, I’m not saying that eventually there will be,


I feel like there will be interventions that are more whole body, but I think the place where we are at the science of aging, I think we will start more with very targeted organ specific interventions like decreasing either muscular schedule aging or decreasing brain aging or decreasing heart aging, so on and so forth. And that’s where kind of these clocks become really, really useful. And then I think a level above that is when, so till now we are looking at these clock


raghav_sehgal (37:04.75)

other. Things get really interesting when you start looking at them in tangent to each other, right, in congruence of each other. And that’s when things start getting really interesting, in my opinion. Things start getting predictive in the sense that now you can start predicting that, hey, you’re in this sort of aging subtype, you’re really predisposed to these XYZ diseases, right? And so you really need to focus, like, this is your risk score of developing this X


years, or this is your score for developing this Y disease in the next 10 years. So let’s really try and prevent that, right? And then the one level above this, and this is where, again, things go a little crazy, and this is more where hopefully at some point we’ll be, is you can not only start, you’re just now looking at data very cross-sectionally. You’re saying that, hey,


hannah_went (37:50.276)



hannah_went (38:03.396)



raghav_sehgal (38:04.61)

data that we have, what we’re really trying to show that, hey, this is your aging score right now, and these are your aging scores, how they’re going to be in the future if you don’t practice any intervention. And these are your aging scores going to be if you practice these interventions. So actually tell, predict that, hey, if you do these things, right now you’re plus two years of age in your hard score. And if you practice these sort of behavioral


raghav_sehgal (38:34.55)

And that’s where things are really, really predictive. They are really able, you’re really able to sort of go into the nitty gritties of that preventative medicine paradigm. And I hope that we get there very soon.


hannah_went (38:51.136)

me too for for sure that would open up so many different doors and windows again for preventative health care and different screening so i think it’s you know the stnmethalation that we’re looking at in this genetic field it just becomes more and more enticing and interesting if we can prevent the onset of disease and push that back and increasing health span of course because i don’t want to live forever if i’m you know disease driven so that’s that’s really the


raghav_sehgal (38:51.37)



raghav_sehgal (39:16.891)

Okay. Okay.


hannah_went (39:21.356)

it seems like this guy is the limit there’s a ton of applications for that system’s aging clock so um i wanted to ask again are there any other reasons why people should look at the system’s aging clock over traditional epi genetic age clocks anything you want to add there i know you mention you know of course you may be you you may be aging at different levels according to that system but anything anything else


raghav_sehgal (39:47.49)

Yeah, I think as I one other point was kind of just the balanced score, right, which is some of these clocks are really trained well to predict certain dimensions of aging and systems age, at least from the attempt that we have made is able to sort of create a more balanced score that is not really biased in one particular dimension. So, you know, what what I’m really saying by that is, see, there’s clock a that’s really good at predicting cardiovascular aging.


and say lung aging or something else. And Clock B is really good at predicting some other type of aging. What’s really happening then, it’s just kind of over predicting for those aging and those dimension and you’re kind of getting the not the most appropriate whole body aging. They’re kind of like overly fit to that one dimension of whole body aging. And that’s not what you really want. You want your whole body aging score, not the aging score of a few dimensions. And so that’s the place


can be truly beneficial. And again, the other thing is kind of this identification of agiotypes, right? I think being able to categorize yourself into agiotypes and knowing that, hey, my agiotype is predisposed to these XYZ diseases. And so those are the ones that I really need to be on the lookout for. It’s again, super insightful. And the other thing that I would say is, and that’s where it kind of


because we have data that allows us to do this. We have done validations that have shown that from about eight years of follow-up, the scores are very predictive of sort of organ-specific disease events. So for example, the cardiovascular clock is very predictive of cardiovascular disease, of myocardial infarction, so on and so forth.


the lung clock is very predictive of lung cancer, the hormone clock of breast cancer, so on and so forth. So what we are really seeing then is that these clocks, given the fact that they are more new and have been validated on larger data sets, are actually potentially, you know, better predictive as well of future disease occurrences. And I think those are all sort of the pros, so to speak, of the systems age framework.


hannah_went (42:17.316)

and the system’s age the system’s age framework really takes the guessing away right right now you receive this biological age score back but you know a lot of times you’re hypothesizing why you may have accelerated aging or accelerated aging and you may be looking at epidemiological correlations and things that have been shown in the research but i think the system’s approach that you have is very unique in the fact that you say ah here’s what’s aging here’s what i need to do i understand my risk


what am i going to do about it right or what is there to o about


raghav_sehgal (42:47.75)

Yeah, yeah, yeah. Yeah, no, exactly. So I think one of the ways I would initially explain this to people is like, hey, you went for a biological age test, whatever kind, not just epigenetic, and they give you a whole body aging score. So what? What do I do with this? How do I intervene? But now, oh, my heart is aging, so maybe I need to practice behavioral interventions that improves my cardiovascular disease, decrease my cardiovascular disease risk, or so on.


hannah_went (43:05.116)



raghav_sehgal (43:18.332)

So now you have a more targeted understanding of how to do this field.


hannah_went (43:24.176)

yeah the system’s age is truly an over arching biological age like this pure aging signal that is fundamental and encompassing of all of these processes so super exciting work well you know roovwerewe’re almost getting to the end here what are you what’s most of your time taking up by right now is it really pushing this this paper you know finishing this up i’m assuming


raghav_sehgal (43:35.492)



raghav_sehgal (43:51.15)

Yeah, no, I think we’re in the final stages of wrapping this paper up. You know, there’s some really cool additional analysis that I really want to do that I’m working on right now, just kind of comparing how some of, you know, one of the things that’s been out there about epigenetic clocks is that people say that, hey, why do we really need epigenetic clocks when we already have these biomarkers, you know, very well validated


hannah_went (44:01.916)



raghav_sehgal (44:21.07)

heart aging. So one of the things that I want to show is that, hey, we’re actually capturing more information than just a simple cholesterol test or a simple, you know, a simple heart rate test or something like that, or some very common scores like the ASCVD score, which is very common in sort of cardiovascular aging. And so if I’m able to show that we can really


raghav_sehgal (44:51.67)

are actually more meaningful and insightful than the present biomarkers out there. And that, in my opinion, would really change the paradigm that we have right now, which people keep on questioning that, hey, okay, you have an epigenetic clock. It’s not as good as everything else that’s out there. But hopefully, if you were able to show that, that would be, I think, a paradigm change.


hannah_went (45:16.716)

i love that and i need that dat it’s hard to convince you know people who have questions about that and it’s great they’re asking the right questions right we need to prove that in show that data you know i’ve heard even some experts say that the den methalatian and epi genetic testing is going to overtake your conventional blood testing or your hormone panel testing one day right because of the insight it can can hopefully show in what you’re working on and not only the insights but identifying those insights or issues


raghav_sehgal (45:17.55)









hannah_went (45:46.336)

earlier again coming back to the therapeutic preventative approach too


raghav_sehgal (45:49.87)

Yeah, and I also feel like just the simplicity of it, honestly, you know, instead of being jabbed like 10 different times for like a metabolic panel and a renal panel and a, you know, and another brain panel or a hermetic panel. Now you’re just being jabbed once and you’re like, Hey, I’m done. I’m good for now. And this, I know this sounds very, uh, perennalsy, but I promise this is not like perennals.


hannah_went (45:54.256)



hannah_went (46:15.616)

people ask us that all the time actually we got to ask that one time when i was on a call and i was wearing my hair up with a black turtle neck and i oh my god i look like elizabeth right now like i promise it was some weird coincidence and i just had to say it yeah yeah now that’s that’s hilarious but yeah i get super excited about the work and and you mentioned yet only one stab instead of all of these different labs and that becomes


raghav_sehgal (46:23.893)

Thank you.


raghav_sehgal (46:31.118)

Promise there are signs backing this up. Promise. There’s valid signs backing this up.


hannah_went (46:45.416)

super you know frustrating and then there’s a lot of education behind all of those labs and then also the cost like you know the cost is going to go way down i haven’t written on my website like if you don’t understand epi genetics don’t worry we didn’t understand genetics that long ago right and now we have the whole human genus sequence you can you know pay less than a hundred dollars to get your entire genu sequence in less than twenty four hours so if you think it’s expensive now the price in is going to really


raghav_sehgal (46:50.591)



raghav_sehgal (47:01.367)



raghav_sehgal (47:10.772)



hannah_went (47:15.576)

crease as as you know yours becomes available


raghav_sehgal (47:16.23)

Yeah. Yeah, no, I think that’s the way forward as we discover new technologies that can process this epigenetic data faster and better and more accurately. The cost is only going to go down. I don’t see a future where maybe, say, five, 10 years from now, it’s just $10 worth of an analysis. And compared to the canonical clinical biomorphists that do,


it’s 200s of dollars at this point.


hannah_went (47:49.536)

definitely definitely well we’ll hope to see that day well you fingers crossed my last question this is when i ask at the end of every single podcast if you could be any animal in the world of what would you be and why


raghav_sehgal (47:51.83)

Fingers crossed.


raghav_sehgal (48:04.23)

If I could be any animal in the world, that’s actually a very interesting question. I think I’d like to be a turtle because they live the longest, at least like live fairly long, 200 years or something. I’d like to see 200 years of the world changing.


hannah_went (48:21.856)



hannah_went (48:25.716)

yeah no one has said a turtle yet so that’s like massive turtles at the zoo and you know here someone zoo keeper speaking you know they’ve lived they’ve been on this earth for like a hundred fifty years and you’re like what and they move so slow and their massive no answer


raghav_sehgal (48:37.65)

Exactly. Yeah, and so I want to see the world change for 200 years. So that would be really cool. Yeah. Fingers crossed.


hannah_went (48:44.976)

yeah and all your work changing the world right


hannah_went (48:50.196)

well perfect we’ve come to the end of this amazing podcast interview i can’t thank you enough and for any listeners who want to connect with you you know where can they find you i’ll put i’ll put everything in the show notes to but where could they connect with you or learn more about your work


raghav_sehgal (49:01.53)

Yeah, well, I’m on LinkedIn, I’m on Twitter. I and of course you can reach me at my email ID. That’s my first name dot second name at Yale dot ETU. And well, if you’re ever in New Haven, just drop me a message. I’ll be more than happy to sort of take you around and talk more about epigenetic cloths.


hannah_went (49:21.956)

awesome well yeah thanks again rob thank you everyone for joining us at the everything epigenetics podcast remember you have control over your epigenetic so tunin next time to learn more thanks again


raghav_sehgal (49:32.611)

Thank you, Hannah. Bye bye.



About this Guest Expert

Raghav Sehgal
Raghav Sehgal is an expert in applying deep learning to multi-omic and multi-granular data to solve aging, with seven years of experience in integrating data science and artificial intelligence in the biomedical field, including building and managing a Series A funded biomedical data science startup.

More About me

Everything epigenetic
Everything epigenetic
MOSAIC: Multi-Organ Scores of Aging across Integrated Components

More Episodes