AI-Powered Longevity Science — One Gene to Reverse Aging? | Daniel Ives PhD

What if an AI could run centuries of aging experiments in a year? Dr. Daniel Ives, CEO of Shift Bioscience, explains how his team used a virtual cell to discover SB000 — a single gene that matches Yamanaka factor rejuvenation without cancer risk.

CHAPTERS:
00:00 — Centuries of experiments in a year
02:03 — Daniel's journey: physics → Aubrey de Grey
10:08 — The epigenetic clock breakthrough
12:09 — The 13 mitochondrial genes
20:09 — Yamanaka factors (OSKM) explained
22:10 — Partial reprogramming: the weekend analogy
24:11 — The cancer risk problem
26:11 — AI virtual cell: how it works
32:12 — AI-driven dark labs
40:16 — Single-gene interventions
42:17 — Shift's discovery: genes that reverse aging
50:19 — Animal testing begins
62:20 — Hearing loss: the unexpected aging connection
66:21 — Rapamycin reverses hearing loss in animals
78:25 — N=1 medicine and wearables
84:26 — Closing

REFERENCES:

Shift Bioscience: shiftbioscience.com

Partial Reprogramming (Nature Comms, 2024): Nature

Epigenetic Clock (Frontiers in Aging, 2024): Frontiers

GUEST: Dr. Daniel Ives, PhD — CEO, Shift Bioscience, Cambridge UK
HOST: Dr. Robert Lufkin MD | robertlufkinmd.com

⭐ Enjoying the show? Please leave a 5-star review on Apple Podcasts — it takes 30 seconds and helps more people discover the science of health and longevity. Thank you!

New episodes every Tuesday & Thursday. Subscribe so you don't miss one.

Continue this conversation on Substack: https://robertlufkinmd.substack.com
Lies I Taught In Medical School — Free sample chapter: https://www.robertlufkinmd.com/lies/

Web: https://www.robertlufkinmd.com
YouTube: https://www.youtube.com/robertlufkinmd
X: https://x.com/robertlufkinmd
Instagram: https://www.instagram.com/robertlufkinmd/
TikTok: https://www.tiktok.com/@robertlufkin
LinkedIn: https://www.linkedin.com/in/robertlufkinmd/

Transcript

A DeLorean For Biology

SPEAKER_00 0:00

Just to try and emphasize how powerful that this is, um, the number of experiments we can do within a year is roughly three centuries of future wet lab experiments. So it's like getting into your DeLorean, going to 2325, getting out the DeLorean, like reading the results, right? Like the best results, coming back to now, right, and doing that in the wet lab. And that's that's sort of the power of like having things on the computer.

SPEAKER_01 0:31

Welcome back to the Health Longevity Secrets Show, where we push the limits of human potential and unlock the secrets to our health and longevity with your host, Dr. Robert Lovkin.

SPEAKER_03 0:42

Today's guest is Dr. Daniel Ives, founder of Shift Bioscience, and his story is a wild pivot from the classic pick a hallmark of aging science to letting the data drive everything. We talked mitochondrial DNA, why his original hypothesis didn't survive the epigenetic clocks, and how Shift now uses single cell aging clocks plus virtual cell models to run literally centuries of experiments in silico. And then it gets even crazier. Single genes that can rejuvenate cells without pluripotency, plus real translational targets like fibrosis and age-related hearing loss. Buckle up. This is the leading edge of the longevity revolution.

SPEAKER_01 1:28

And now, please enjoy this week's episode.

SPEAKER_03 1:34

Hey Daniel, welcome to the program.

SPEAKER_00 1:38

Great to be on, Rob. Great to be on. I appreciate the invitation.

SPEAKER_03 1:42

I'm so excited to talk with you today. You know, you you you have an amazing journey from studying mitochondrial DNA mutations at Cambridge to founding really one of the leading companies in the longevity space shift bioscience. And and before we dive into that, let's maybe take a moment and tell us a little bit about your journey, how you you came from where you started to hear.

SPEAKER_00 2:11

Yeah, so this is this is I think a common story. Um I watched a TED talk by this scientist called Aubrey de Grey and just captured my imagination, you know, just such a spectacular talk around the time I was graduating. And uh he you know he said two things. Firstly, let's do something about aging, right? Let's not just assume it's off limits, right, to intervention, like don't put it on this pedestal. Let's get off our you know, get off our seat, start doing something. And then immediately afterwards, which is key, he said, here's a framework for aging, like breaking it down into digestible pieces, which sort of became the became the hallmarks of aging, right? This framework, and you know, you just list them out and he says, you know, here's an order of hallmarks. And I I bought his book immediately, read the book. I think in the book it said the mitochondrial genome is one of the hardest things to solve. So I was like, hell yes, that's what I'm gonna try and do. I'm gonna try and solve this genome. And uh he'd sponsored a uh PhD um in the University of Cambridge in the UK, and uh so I I sort of reached out to the PI that was collaborating with Aubrey, and unfortunately that project had been shut down. Um I didn't have a plan B, and then I found a plan B within the lab and got to work trying to solve aging, and I I believed what I was working on was a sole cause of aging and it would fix it in its entirety. So when I did solve this tiny little problem, I had that feeling like I'd solved aging and uh after that I was determined to take that to the world um and eventually that became shift bioscience. And the first thing we did as a company was disprove the hypothesis behind the company. So we we showed this little thing, which is just a little thing. I mean maybe it will be a bigger thing in the future, but it was one part of the picture and uh ended up pivoting the company hard into machine learning as the route to the biology that we want to manipulate to have the biggest effect size on things like lifespan eventually. Um yeah, and along the way, just yeah, spending my own money, spending other people's money, um yeah, eventually bring an investor on board, staying very little, um, then getting very big very quickly, and just all of the you know, all of the uh faculties you have to develop to sort of manage a team at a higher level, having to communicate, right, having to learn how to do that. Um but yeah, it's it's a it's a pretty crazy journey trying to confront this thing. It's like it's it's like a perpetual learning experience, right? The the more you see of it, the more you realize is ahead, and you don't want to let that put you off, right? You want to reel in the finishing line every opportunity you can, but then you've got to try and kill what you're doing all of the time, not to waste time. Um, so I just hope I'm I survive the process.

What Aging Means Without Mechanisms

SPEAKER_03 5:09

That's great, great, great introduction. It's funny, I I I didn't realize that Aubrey's talk influenced you. It it's what a coincidence because I was I was up in Monterey and I was going to this new conference, or it wasn't new, it was new to me. It was called TED, and they uh they were they had these speakers, and I remember sitting in the audience when Aubrey gave that talk, and it was it was a great talk, it was mind-blowing. And and it's so funny you you uh you you listened to it as well. I mean that that's a great conference. I think the same conference or one of the one of the two of the years before or after, they had another guy from uh he was a guy who'd started PayPal, and he was he had an electric car that he brought there that he was showing everybody named Elon Musk, and he was trying to launch this Tesla thing. But but that's fascinating with with Aubrey. He's such such an inspiration, and it's it's amazing to think that you, yeah, that that he's influenced the course of of your career that way, and now you're you're uh taking up the the torch and going on. Uh of course, aubrey's still doing great things too, but uh it's nice to see that. Um well, I'm really looking forward to getting into a little bit of details on this, and I don't want to lose some of our audience who may not know some of the technical stuff. So I let's just start with at the beginning, kind of define some of the terms what we agree on, like like aging and longevity. So, like what what is your how do you think about longevity and aging? You know, is it is it just oh, our cells wear out and it's just inexorable time, or is it you know hyperfunction like Black Lascone and you know and others, or is it is it something else? What what why do we age? What causes aging? What are we what are we trying to fix here?

SPEAKER_00 7:11

Yeah, so I'm sort of shaped by my personal experience, which is I sort of dived hard into hypothesis-based science, you know. So I, you know, Aubrey, hallmarks of aging. I chose the hallmark I think was most causative, right, based on the evidence. I worked on that very diligently, and then I sort of, you know, I tried to go as far as I can and ended up disproving um this was a causal factor, right, at least with the technology available. So I ended up basically hanging up my boots as a hypothesis-driven scientist and embracing the world of unbiased science where you use very rich signatures like single-cell transcriptomics or any types of omics, and you let the data take you to the answer, because you know that the cell is just this unbelievable physical computer, right? Like, and you know, if you if you've done biochemistry biology, it's like a Lord of the Rings epic. Like you read the first chapter, it's just like immunology, and then there's like basic biochemistry, and it's like you're you're you're learning all these characters and how they behave, and that's just what we know about the characters and how they interact, and there's there's everything that remains. So the idea that you could just understand a part of that infinite complexity, sort of you know, I think what tends to happen is you create a comfort zone of mechanism that's still surrounded by infinite black box. I sort of found out the hard way because I was you know risking my own money and time. So I fully embraced unbiased science, and that's you know, now it's machine learning. So I I'm gonna give you a really terrible answer, which is I I don't actually worry about exactly how it's happening. We just have certain metrics for being older, being younger, which are unbiased metrics, and we we sort of we they they're the Lord, right? The metrics are the Lord, and they might lead us down a dead end, right? And we we fully acknowledge that, and that's a risk, but ultimately they're like a compass that we're so lucky to have. And so for me, I think what what will happen is we'll find this biology that does this remarkable stuff, and it'll be like finding a UFO in your backyard, and then we'll have to reverse engineer it, and that will reveal the elegance and the majesty of the rejuvenation process. But that's the way around I see it. Maybe when we get close to the answer, we'll start, you know, mechanism will start catching up. But I'm not putting mechanism in the way of utility, right? So I just want to get to utility and let the machine learning take us there, and then we can enjoy the story at the end. Um, I don't know if that's the correct approach or not.

SPEAKER_03 9:41

I love that. I love that. Sort of, I think we could all benefit from taking that strategy to other things we do in our lives and other approaches to things. In other words, look at you know, almost first principles and and look at the what the evidence produces and then have the hypothesis afterwards. And and really, you know, it sort of turns the science a little bit on its head. You know, back in the day, we we have to be hypothesis driven because you know, uh, doing one experiment was a big deal, took a lot of time and money. But now with AI, we can literally generate you know many hypotheses in parallel and and basically let mechanisms drop out of that. Right?

SPEAKER_00 10:21

Yeah, just just to just to add sort of add one line on top of that, which is with these systems, we we we can actually do every experiment we want to do. Right, this is the great thing. We don't have to we don't have to be choosy, right? Like we, you know, we've only got a certain amount of time, so we can only choose one experiment, and then we go, we can actually we've got such experimental power now that we can effectively do every experiment and then just choose the best results to take into the wet lab, and then we do, you know, we basically just confirm on on the back end. So um that this sort of stuff wasn't possible until very recently. It's really a product of the sort of the compute and the models that sit inside that compute. Um, so you know, I'm very lucky that this this has happened around the time I was trying to solve this problem, right? Without without these capabilities, we'd be struggling away for like many many moons, um, but we're able to find stuff much sooner.

Mitochondrial DNA Hypothesis Meets Clocks

SPEAKER_03 11:18

Yeah, and we're and we're gonna dive into that at you know when we talk about specifically what you're doing at Shift Bio um science. But before before we leave this, maybe what was the failed hypothesis that you tested? It was around mitochondria, right? Uh that that your initial thing in grad school, I guess.

SPEAKER_00 11:38

Yeah, so um sort of basic biology of the eukaryotic cell. You know, we we are sort of multicellular version of the eukaryotic cell, but each cell has two genomes. So there's there's the famous genome, which is the nucleus, and there's the less less recognized genome, which is the mitochondrial DNA. And over time, um, most of the genes from the mitochondrial DNA have have sort of migrated west right into the nucleus. They found a safer home in the nucleus, but these 13 genes that make 13 proteins persist in the mitochondria and they're encoded on the mitochondrial DNA. And it's actually a whole genetic system separate to the nuclear genetic system. It's like running an Android, well, it's like running a Mac inside a Windows operating system, right? It's like sort of bizarre, and it's very expensive to do this as well. Like you need a whole set of new uh ribosomes and tRNAs to run the system, um, but that genome's very important, and uh the hypothesis I was working with was mutations in this genome are the underlying driver of aging, and there's some very provocative evidence that this was the case. So, for instance, if you delete the proofreading domain in the polymerase that reproduces this circular genome, you basically create about two and a half thousand times the level of damage every time you replicate the genome. So it's like it's like just you know firing damage into this tiny little genome, and you create a very comprehensive human aging phenotype in a mouse. So this mouse loses hair behind its neck, gets kyphosis of the spine, it gets enlarged heart. This is actually more like a human aging phenotype rather than a mouse aging phenotype. And yeah, I think there was like a nature front page at the time, like mutator mouse aging solved. Like it was you know, paraphrase, right? They were like, We solved aging. And I'm reading that and I'm like, hell yeah, I'd already chosen that, and it's confirmation biased. Um, so you know, we got hold of one of these mice, and I got you know, I had this type of therapeutic, and we put it in the mouse, and then we managed to slow aging and partly reverse aspects of aging in this mouse, only confirming my belief I'd solved aging. Um, but this all came to a screeching halt with um the well the discovery of the epigenetic aging clock. So Steve Horvath that you know discovered this very robust multi-tissue biomarker of aging within the methylome. Uh he benefited from these huge databases full of omic data like methylome data. Um but the these clocks were like the major breakthrough for the field, like having a measure of aging that you can experiment against instead of having to run huge lifespan experiments. It was like it was like a really big step forwards. So I wanted to use this clock to prove to the world that damage to the mitochondrial genome was the major cause of aging, and by affecting that damage, I would you know stop slow reverse aging. And what I showed was that by changing this damage had it had zero effect on the clock. So the clock was like blase, right? Like we we put the damage up, it didn't it didn't spin faster, we put the damage down, you know, it didn't spin slower. So, you know, this damage was creating an aging phenotype, but that aging phenotype wasn't driven by an acceleration of physiological aging, it was driven by a genetic change, specifically the mitochondrial genome, that has the common phenotype with physiological aging. So I had a lot of faith in this biomarker. Like I was hell bent that this biomarker was the compass we need to follow. So for me, you know, if we didn't get the answer, that was a reason to move away from this flawed approach and do something else, and that's exactly what happened. So the results came in and uh we basically fully embraced aging clocks as the route forwards, and we basically you know we we we sort of doubled down on clocks, make them high throughput, do CRISPR screens where we can scan all biology, looking at the clock, seeing where it's responding. Um, and there's you know, so basically a supercharged version of that, which we can do in the computer now, run all experiments, see what the clock's doing, bring the bring the winners into the wet lab, and that's that's how we find the genes.

SPEAKER_03 15:50

Well, and put let's put a pin in the clocks too, because I want to come back and talk about clocks when we talk about shift bio and specifically what you're doing, and that's that's so important. That's fascinating because like the there's some people believe that uh with mitochondrial DNA mutations and impeded energy metabolism, right? That you know, the cell just can't do that. The impeded energy metabolism drives epigenetic changes on the on the epigenome in the nuclear DNA, forcing it to select more energy efficient functions. I mean, it just makes sense. Um, you know, if you can't operate, if you're not making energy, you'll select, you know, you won't do certain things, but you're saying that didn't really pan out much that there weren't epigenetic signatures that corresponded to the mitochondrial DNA changes, really, right?

SPEAKER_00 16:42

Yeah, so so within the narrow window of the mitochondrial DNA, and basically this, you know, uh, I think it's you know, two and a half thousand times more damage. We didn't see the clock respond, but then within a broader window, which is like mitochondrial numbers or mitochondrial function, um, there is a relationship. So Ken Raj, who's now at Altos lab, he he published a study which was assessing all of the different hallmarks of aging and how they affected the clock, right? He wanted to sort of systematically uh change those hallmarks in you know the positive and negative direction and see which hallmarks were driving this clock, maybe um yeah, yeah, like basically which hallmarks are more important, and he did find a relationship with mitochondria, and it was a reciprocal relationship. So I think he used besofibrates to boost the biogenesis of the mitochondria, and the clock ticking slowed down, so it sort of slowed the rate of aging or epigenetic aging. He then used another reagent which is called uh CCCP, which is basically it depolarizes the inner mitochondrial membrane, and this this basically interrupts function and it made the clock spin faster. So you have this reciprocal relationship, right? Like make more mitochondria, there's more mitochondrial function, clock the clock ticks slower, disrupt mitochondrial function, the clock spins faster. So within the narrow window that I was asking my question, which was just mutations, there wasn't a relationship. If you just go a little bit broader, there was a relationship. So it's just it just shows you that sometimes you miss the bigger picture and you make a conclusion that's a little bit too, you know, yeah, it's it's ultimately a false conclusion. There is mitochondrial uh mutations in isolation didn't have a link to the clock, but then when you consider mitochondrial numbers that did have a relationship. So maybe there's a threshold of head with mutations where they get bad enough, starts impacting function like CCCP, and then you'd see a relationship. We just don't we didn't encounter that in these mouse experiments we ran initially.

SPEAKER_03 18:44

Oh, okay. So the the pure mitochondrial mutation didn't show that association, but it and that might be that might be due to cell uh basically number size, the quantity of the mutated ones or the percentage of the mutated ones, and all because I'm just thinking of you know uh other people in the mitochondrial transplantation space, like you know, our mutual friend Tom at Matrix Bio and Natalie at Menovia and several other people they've all been on our podcast talking about mitochondrial transplantation as a as a rejuvenating theory. So you there's still maybe uh you you're not discounting that it's just purely the mutations itself as a as an effect, right?

SPEAKER_00 19:28

Yeah, there's there's a basis, there's a basis for increasing the number of mitochondria as a means to slow down the ticking of the clock. So that that is one of the findings of uh the this paper from Kanraj. If you boost the numbers of mitochondria, you can slow down the rate of aging according to these clocks. Interesting. I think mitochondria is certainly part of the picture, um, but more at the macro level, not at the micro level of of the genome, at least you know, within the boundaries of the experiment we did.

Partial Reprogramming And Its Tradeoffs

SPEAKER_03 19:54

Yeah, yeah. So okay, well, good, not purely the mutation. Okay, so then shift over one other thing we're gonna talk. I wanted to talk about was partial epigenetic reprogramming, which is in the news with you know, so the classical approach with Yamanaka factors. Um what is that exactly? And um what are the and more what are the limitations of that approach uh compared to what what you're what you're doing?

SPEAKER_00 20:23

Yeah, so start with your first question. What is it? Like what is uh partial reprogramming? So we we know precisely zero about it, except there are four factors called the Amanaka factors, they're transcription factors. When you express four of them all together, they can turn adult cells into stem cells, and in the process they reverse age by almost every measure, including the epigenetic clock. You can you can substitute, well you can minus CMIC, which is one of the four factors that's an oncogene, and you can maintain rejuvenation even with a subset. Um but the biology that sits underneath these factors in terms of rejuvenation is a mystery, right? It requires um it requires biology to be mapped in a way that's uh quite difficult to do. So yeah, I mean what's wrong with these factors? Maybe let's just start with that, right? What's wrong with them? Um it's actually what was right with them. That's the thing. What's wrong with them is what was right with them. So what was the original experiment? Yamanaka was trying to m turn adult cells into stem cells because embryonic stem cells were not uh readily available to researchers, so the idea you could just make them right from a readily available source was like that was a real research area that could uh do a lot of good. And so he he fully optimized his his experiment to find genes that make adult cells into stem cells. That's that's the feature he was optimizing. The bug was rejuvenation. So the rejuvenation was discovered quite quite a while later. And that is the bug, right? That remains the bug of the Yamunako factors, not the feature. So you can try and optimize for the bug, which is called partial reprogramming. Let's just use it for two days at the weekends, right? Let's use it for the weekends. And we can get we can actually extract a positive effect out of these factors. Like there's that's the therapeutic window. And there's certain use cases where you know there has been a safety case that's been built around using the factors. Uh I think in the the eye, certainly the eye is one example. Um I think Yuri Dygon from Youth Bio, there's that's sort of a case for the brain, it's non-proliferative, so it's not as dangerous to factors. Um, but ultimately these factors we're we're still optimizing the bug when we're talking about rejuvenation. So the the interesting question is like, what is the rejuvenation biology? Because if we could learn about that, we could double down on the bug, you know, optimize for the bug, like directly optimize for rejuvenation instead of just optimize for the bug, which is very hard to do. Um, so that's that's you know, ripe for discovery. Like, how do we how do we approach that problem? Like, which tools do we need? Um, you know, what do we find? Like, do we find anything that's better than amalaca factors? Does it need to be combinations? You know, could it could it be inhibition? Does it have to be overexpression? Like, there's this whole world that needs to be explored, and so you know, we we explore this world, we're finding some very interesting things. Others are exploring this world, they're also uh doing great work. Um, but the great news is there seems to be biology beyond the amalaca factors that's involved in cell rejuvenation.

SPEAKER_03 23:37

So even the and I I think we're all aware of uh Dave Sinclair's work with the the trial going on in Boston, like you say, in the with the uh certain types of age-related glaucoma and nion in humans, the first clinical trial. And they're and they're using OSK, right? The subset of Yamanaka factors, and hoping to reduce the pluripotency effect of that and the you know the the teratomas and the all that things. But you're saying that that really doesn't really doesn't do it. And you're we're gonna talk about in a minute your approach, which which um is really interesting from again first principles kind of from the AI and how that drives it, right?

SPEAKER_00 24:20

Yeah, so I think I think David does have a case. He does have a case in the eye. So OSK can be used with a broader therapeutic window than the four four factors together. And I you know, my understanding is they built a sufficient safety case around this use case in the eye. Um, so I think there's a good chance they do something very positive. Um my only my only concern is that they're still using factors that were optimized to be pluripotent-inducing factors, right? They were designed to make stem cells. So I think they they you know you can optimize for the bug and get something out of that. Um but these genes were always designed to do something else. That's my only bugbear, right, with with the uh use of Yamanako factors. I'm excited because it's the first trial using cell rejuvenation as the mechanism of action. Like it's the first one to sort of break into clinical development. And yeah, that that's that's a great thing, right? It's you know, it's it's a long time coming. Um, but I don't I don't think that's the I don't think that's the iPhone 10. I think that's the iPhone 1. And the iPhone 1 was never designed to be an iPhone, right? It was sort of massaged, maybe it was like a power tool, right? And it was like adapted into a phone. So, you know, I I I think it's good it's happening, but there's a lot more we can do just by understanding what rejuvenation is, and you know, iPhone 2, 3, 4, 5, I think there's a lot of headroom uh to do much, much more spectacular things.

Single Cell Clocks And Virtual Cells

SPEAKER_03 25:49

Yeah, we're we're we're just at the at the cusp of this revolution, and you know, it's it's such an exciting time to be alive. I there's no other time in history, I think, like this, and certainly uh no other time I'd want to be alive. It's it's every day is uh new stuff happening. So so let's let's pivot to to AI and talk a little bit about that. And you know, everyone talks about AI, the revolution, it's happening, you know, uh gold, you know, uh math Olympiad gold medals are now you know owned by AI, and the you know, the benchmarks of these frontier labs keep getting, you know, keep getting moved and moved and moved, you know, eventually in many areas already surpassing humans, but it's it's going on. So at Shift Bioscience, you're using machine learning to uncover these true drivers of aging. So can you can you break that down a little bit and how you approach it? It's fascinating.

SPEAKER_00 26:45

Yeah, so we we we use narrow AI. So this is you know, by design, looking at sort of a very narrow problem, and we're like puddling that problem with GPUs, right? It's like you just keep throwing the GPUs into the problem. Um so so the narrow AI, there's two there's two types of narrow AI that we're working on, and one is an aging clock. So the original aging clocks were actually one of the early fruits of machine learning applied to biology, like there was it's basically been impossible to find a biomarker of aging that's high accuracy and works across tissues. Um, but with them within the methylome, if you if you take all that methylome data with age labels and you use relatively simple machine learning, it's not like fancy transformer models or you know graph neural nets, just simple simple linear models, um, you can train a biomarker that's highly correlated across multiple tissues. So that's the first narrow AI. So there was the original clock, we've gone one step further, which is a single cell aging clock that basically brings the epigenetic clock into the single cell world via gene expression. So there's a bit of complexity here, which is you've got the epigenetic clock, which can hear aging, doesn't tell us what aging is, but can hear aging, right? You've got all these methylome sites changing, and it gives you like the sense of aging. And then what we've done is we've from the gene's gene expression or the gene signature, we can hear the epigenetic clock. So the epigenetic clock hears aging, and we we hear the epigenetic clock from gene expression. So it's you know, uh for lack of a better expression, it's like Chinese whispers. So, you know, epigenetic clock hears aging and we hear epigenetic clock, and the message might be slightly distorted, but it does give us high throughput uh exploration on the aging problem, right? You you're able to do like single cell CRISPR screens and see which you know which genes are affecting aging. Um and that that brings me on to the second narrow AI tool that we've developed. So this isn't our sole development, this is something we've fine-tuned um from the literature. So there's these open source virtual cell models. So I'm gonna give you a very precise definition of what virtual cell means to us. What it means is um a representation of how genes impact each other. And the way that this is created is you feed quite advanced machine learning models now like Transformers and Graft NeuralNets, you feed them loads of single-cell gene expression data. So that these models look at lots and lots of cells, they look at what the genes are doing, and they basically stitch together. If gene A is up, gene B is down, if gene D is up, gene F and Z are down. So it can, you know, from the ground up establish how genes interact with each other. You can put that in the computer, and now you can effectively run uh virtual experiments. So I can say I'm gonna overexpress two genes or four genes like the amino acofactors, what does that do to the rest of the genes in the genome? And because we have another narrow AI tool, which is a single cell aging clock, we can convert what's happened in that in silico experiment into an age label. So we can we can say what was the age at the beginning, what was the age at the end, and so we can say like what was rejuvenative out of you know, were these two genes rejuvenative when we overexpressed them? Were these four genes rejuvenative? Like how more rejuvenative was one experiment versus the other? So you can see we can do a lot of stuff all of a sudden, and I I think just to try and emphasize how powerful that this is, um, the number of experiments we can do within a year is roughly three centuries of future wet lab experiments. So it's like getting into your DeLorean, going to 2325, getting out the DeLorean, like reading the results, right? Like the best results, coming back to now, right, and doing that in the wet lab. And that's that's sort of the power of like having things on the computer, uh having these uh aging biomarkers you can use in these types of situations. So these are two narrow AI tools, right? Single cell aging clock, virtual cell. We use one inside the other. I call it like the the dance of the black boxes. Um, and then yeah, just very very long answer, sorry, but then a last AI tool, and this is something we're not deploying right now, but it's this idea of the AI taking over from the scientist, like sort of taking over decision making, uh deciding which experiments to do. Um sorts I think it's like agentic AI, right? Like sort of making a better scientist, like an AI scientist. So I think we're not quite there yet in terms of like you know, we're we're we're at the mercy of that decision maker, but it's certainly getting more powerful every day. And this is where you know when people talk about AGI and superintelligence, you know, imagine that in combination with these narrow AI tools, like the amount of you know iteration that's gonna be happening um relatively quickly. So you know, we're already finding things with our mushy biological brains in the mix with the narrow AI. Uh so at some point, I don't know when that is. We we prepare for that, we basically prepare as if that's never gonna happen. Um, but if that does happen, that's gonna be a huge supercharger as well.

SPEAKER_03 32:04

Well, yeah, I'm I remember the the quote that some of the people are making now about scientific discovery powered by AI, and some are even predicting that you know, in the next 10 years we're gonna have more scientific discovery than we've had in the last hundred years, you know, based on the power of AI to drive to drive innovation and discovery. And and it, you know, since you mentioned that Steve Horvath was at UCLA, which is just down the street from where I'm sitting here today, when he, you know, when he did the original uh methylation clocks back in the day. Now he's up in Altos and uh in the Bay Area. But uh but that work there now where does that fit into uh I assume these are you're not using LLMs or even deep convolutional neural networks? It's back to is it deep learning at all, or is it more what what are the what type of AI are you using on these so the the answer is all of the types?

SPEAKER_00 33:05

So we we we basically we have all the data and then we we try all of the different models and we basically just look at the result like which model creates the best single cell aging clock. And I think the answer I can give you is it's actually a mix of a lot of things, right? It's like it's you know it's like it's like you you need to be a good chef, right? You just need to try and um you need to try a lot of things, right? There's obviously lots of things that aren't going to work, but we we mix some of the models together. That's also true for the virtual cell. So um I guess one of the models we high hold in high regard is called SCGPT, so it's called single cell GPT, and that's a transformer model that sits underneath. Um there's a graphed neural net version of the virtual cell, I think it's called Gears, and there's a more recent one called State. So there's all sorts of models, and we again we just stay agnostic to model, we just try all of the models and see which ones are most accurate, which ones are most robust. Um so yeah, it's it's it's it's a sort of a strange process, right? Because you just try everything and then you choose the best thing. It it's almost like how evolution operates. It's sort of it's not thinking, right? You're just you're thinking enough, right? You're like, here's a bunch of var variety that we want to test, and you you're gonna source that, but then ultimately you're just gonna see like like hunger games, who who lives, who dies, or you know, it's it's more of a hierarchy, right? Versus life and death. Um, but yes, there's that there's not that much putting your brain in the mix apart from just trying to steer this in the right way, keeping your eyes open. Um yeah, I think I think that's did that answer the question, Robert.

SPEAKER_03 34:40

Yeah, yeah. I mean it's it's almost it's it's what we're hearing more and more with sort of the the modern deep learning paradigm where we don't really program the the computer with the task, we just program it with the the tools, you know, the the the convolutional neural network, the LLM, and then it learns from the data, and the data is domain specific. So, you know, it to some extent you don't even have to. I mean, obviously, you guys are domain experts and you're adding a lot of value there, but a lot of it is the machine. It's it's fascinating. I'm thinking of um go ahead, yeah.

SPEAKER_00 35:17

Yeah, so I think the definition of expert has changed very quickly. So in the world of machine learning, expert means the data set that you have. So, did you build an expansive data set with high quality control on the data? Because that's gonna give you the expertise. If you have a messy data, you know, if you have a messy data set, poor quality data set, you can't be an expert with all of that noise. So, yeah, machine learning, like the data set conquers all. Um, with a great data set, you can actually use pretty primitive machine learning to reach your outcome. Um, but then if you have a great data set and a great model, right, you can hit new heights. If you have a bad data set, it doesn't matter how good the model is on top of that, you're just gonna be hitting false positives and negatives, you're gonna be, you know, you're gonna be stuck in the noise. So, yeah, that's that's our I guess our definition of expert is not a human definition anymore. It's uh it's a sort of machine learning definition, which is like the data set. I think you know there's there's many examples of this. So um bioage, like where's their expertise come from? It's this big data set that sits underneath, right? They have this sort of uh biobank data that they have an exclusive license to, and uh many of these sort of really exciting new uh yeah, machine learning in biology companies, it's all about building these huge data sets. So uh yeah, the human brain is out of the mix, right? It's just sitting there in the data, and the machine learning can see things that are just like many, many levels um below the surface.

SPEAKER_03 36:43

Yeah, and it it the it brings me to the you kind of your third step there and reminds me of um some work. George Church is a friend, and he's a scientist, he's a geneticist based out of Harvard. You know, he famously founded Colossal Labs, you know, to bring back the woolly mammoth and you know, other, but he's done a lot of other stuff. But one of the things he's working on right now that is amazing is the is the Lila Labs, which is uh a dark lab that uh does AI generated hypothesis, AI generated testing, and then even the AI-driven robots that do the pipetting and the smearing and the feeding the mice, and it's called a dark lab because there are no lights, because they're no humans, and it runs 24 hours a day continuously testing and innovating and re-hypothesizing kind of what you're doing. And they've they've just raised$500 million for this project. Yeah, and they're they're one of several dark labs that are you know going this way. There's even there's even a dark uh dark automobile maker in uh China that's the dark factory. I'm blanking on the name, but I'm sure there'll be more, you know. But uh and that's and that's sort of your next phase, or are you already you're already doing uh heavy AI automation? It sounds like you're doing it with the the hypothesis testing. Are you doing it with the actual wet lab part as well? Have you gotten there yet?

SPEAKER_00 38:13

Yeah, so these sort of dark labs where you have the whole scientific process, that's that's that's how you like how do you iterate as fast as possible, right? It's those sort of setups. It's like I'm gonna test something and then I'm gonna see what the real world, you know, I've got an idea, what does the real world say, and then I'm gonna use that information to feed, like update my model and just keep going around until distance between real world and my model is minimum. Um we've got our setup's slightly um different, so it's it's more that we we just do every experiment. So there is iteration, there is iteration, but it's actually um relatively infrequent. It's we we just build tools, we do every experiment, we end up you know enriching the signal, right? You just take the extremes of the distribution and we take that into the real world, and there may be some error, and we can use that to update our models, but it's mostly just throughput. When you when you put the the the laboratory in the computer, so this is not a this is I mean inside the computer is dark as well. Maybe there's a few sparks, right, as the as the electrons move around. But like I think that the better the better description of what we're doing is like the whole lab's in the computer, like we're just doing every experiment in the computer and just taking the best things to the real world, and then we can update our systems once in a while, right, based on all that data that we we generate in the wet lab. But the great irony of our wet lab is that most of the wet lab scientists are just generating the data set to make a better model. So, you know, most of the expense of the company is simply data set and managing that properly, and then occasionally we get these predictions from you know this machine learning in in the cloud, in the cloud lab, and we bring that home, and the scientists will test that, and we're getting great conversion rates. So, yeah, most yeah, we basically we feed this model and then we see what it does and digest, and then it gives us it sort of throws us a crumb, which is this prediction, then we test that. Um, so it's slightly different to these dark labs, right? It's not it's not quite as sci-fi. Um, but actually I'd argue it's sort of more sci-fi. More sci-fi.

SPEAKER_03 40:22

Yeah. Sorry, didn't mean to interrupt.

SPEAKER_00 40:26

Yeah. But it works, that's the thing. It actually works. So, you know, it sort of sounds exciting, but it if it doesn't have a utility, like it doesn't matter if it's exciting, but it actually works. Like, that's the amazing thing. So um these all of these tools were you know, we were trying them for the first time, we didn't even know if they would do anything useful, but we sort of put them together and they did something useful, and now you can iterate and we've got much more powerful tools. And the question now for us is did we harvest everything the first time around? Is there anything left the the second time, right? If we if we gear up with better tools, are we gonna find as much new biology again, or are we only only gonna find like an increment, like an extra 20%? So, yeah, that's that's our next step. Like when we turn the handle of the machine the second time, like what's new and what's already been seen, and that's gonna give us an idea of like it have we harvested everything, and now is it time to move on, like fully double down on progressing biology to different sort of therapeutics?

Single Genes That Beat OSK

SPEAKER_03 41:25

Yeah, yeah. Well, well, pivoting now to to shift biosciences and particularly what you what you're doing with with the with the Yamanaka factor-driven partial epigenetic reprogramming, right? When Shinji Yamanaka did that, he had these transcription factors that activated a whole bunch of genes that um that are presumably related to embryogenesis, right? To to the natural thing that happens when when our cells uh reprogram our epigenome, which happens every generation, right? Otherwise, we wouldn't be there. So somehow tied to that. Now, your approach is I mean, it's different in several ways, but you're also targeting, you're identifying individual genes that either accelerate aging or or slow down aging, right? And and um so so much fewer, fewer genes. Do you see, do you see yourself um, I mean, is is it is an advantage, I guess, computationally, and you know, you can answer that, but do you do you see yourself as doing like a symphony of genes, kind of like the Amonaka factors do, sort of a polygenic effect? I mean, is there a disadvantage if you have to do it gene by gene by gene versus the good news is you know exactly what genes are, whereas the Yamanaka factors you don't necessarily know, correct?

SPEAKER_00 42:49

Yeah, so we expected when we tried to search for you know things that weren't Yamanaka factors that it would be another combination. Yeah, that was that was our um assumption, right? There are four Yamanaka factors, maybe you can get away with three Yamanaka factors. Um, but if we're gonna find, you know, if we're gonna find anything better, it's also gonna look like that, right? Like that makes sense. Like, what do we know about rejuvenation? Just Yamanaka factors. So you you know, you don't want to stray stray too far from you know far you know far from home. Um so we geared up to explore combinations, right? That was having a single cell aging clock and having a virtual cell, so we can explore combinatorial space. So um just for an you know, just just just an example, um, if we want to test all combinations of free transcription factors, which is sort of the minimal subset of the Amanaka factors that is rejuvenative, that's 562 million experiments. And we are we are talking like centuries of wet lab experiments to do that. And so that necessitates having a virtual cell sort of you know run um run a a meaningful exploration, right? If you only sort of it's like battleship, right? If you only put a few pegs down, you're never gonna hit the ship. And if The ship is 562 million holes. Yeah. You only got like you only got three shots. Yeah, it's it's like you know, it's it's impossible. So so with these tools, we can actually put a peg in every single hole. That's the amazing thing. So so this is this is the thing. We geared up for combinations. The first thing we discovered was we there were single genes that were common to combinations predicted to be rejuvenative by the virtual cell. So this gene kept cropping up. And like we you know, we're just looking at it, like because we still get to look at the results, right? As you know, we don't abandon the science entirely. Well, we just look at things and just what you can notice. This is like basic recognition, right? Oh, there's a gene, and it's you know it's coming up multiple times. And so we didn't believe it could be that easy, right? That here's four to three to four Yaman accofactors, and now here's a single gene that's potentially driving all these phenomena. And so we just speculatively took it into the wet lab. We actually just went to the gold standard uh for us epigenetic clock assays immediately. They're quite low throughput, even though it's better than lyspan, it's still low throughput. And huge surprise, uh, the single gene rejuvenated multiple cell types, it rejuvenated more than the Amelaka factors, roughly twice the velocity of rejuvenation, and no pluripotency. So yeah, we were sort of we were sort of we were doing like a double take. We're like, no, no, single gene, like let's try it again, it's the same result. Um, let's try all the clocks, right? Maybe it was the first clock, and then we say, No, it's it works across all the clocks, do the pluripotency assay again. And actually, when you think about it, like your assumptions going in, they colour your conclusions. So yeah, people were like, like, you know, it's not possible, but then the only thing we knew about rejuvenation beforehand was Yamanaka Father. So, like, but the definition of rejuvenation was pluripotency. So, like it you shouldn't be that surprised that if you directly optimize for rejuvenation, you don't necessarily bring across an entirely different pathway that's yeah, yeah, that changes cell identity. So, um the irony of this story is that we geared up for combinations, we built all of these machine learning tools, and the first thing it told us was there are single genes where we don't even need these tools anymore. Yeah, you could just systematically go through single genes in the wet lab without a virtual cell, and that's what we did as a second step. And then the big surprise was how many single genes are influencing the aging process. So we ran a poll in in the company, right? Like everybody had to guess, and whoever was closest won the prize. I don't remember what the price was, something very modest, yeah. And you know, I I think I put down 10. I was like, 10 out of the 1500 genes we're gonna test will be uh you know involved in aging. Uh some crazy guy put 150. He wasn't crazy enough, it was 190 genes, so it was like roughly 10% of these 1500 genes were impacting age. About 150 was in the rejuvenative direction, so the the majority of the 190, but 40 were in the pro-aging direction. Uh, and amongst that we saw Yamanaka factors, um, the combination of the factors, because we wanted to use that as a control. The individual factors didn't actually feature very highly, but then we find all these single genes that some of which outperform the combination of factors, um, and this was a huge surprise, just the sheer volume of biology impacting uh these processes, and then um yeah, just the fact they're single genes. So, yeah, I guess the the funnel, right? Virtual cell, I think there was like uh 350,000 combinations of two. Then we took 1500 into the wet lab. From that 1500, we found 190 with a relationship. We took the top 50 from that broad screen into epigenetic clock assays, and about 30 of those genes uh reverse epigenetic aging, 10 of which outperform Yamanaco factors. So you sort of see the numbers. Um and just just last point on this single gene that we discovered initially in the um simulation. Um, it's actually part of a family. So the transcription factor family, multiple family members are rejuvenative, but one of the family members has no rejuvenation. So you've got very similar protein sequences, and then you've got this one member, again a similar protein sequence, but no rejuvenation whatsoever. So it gives us this sort of window into mechanism because you can find what's the specific sequence on the rejuvenating family members that's absent in the non-rejuvenating family member, and you can start like co-folding other proteins to that sequence, and that suggests you know other candidates that might be mediating rejuvenation downstream of the transcription factors. So, um, yeah, this gene was the most rejuvenative, but then it revealed a family, so it's not just some random gene, right? There seems to be a cluster of rejuvenation genes. So it feels to me like we're getting closer to pathways and mechanism and things like that. Um, but yeah, just having the tools to do a broader screen optimizing for rejuvenation, um, the amount of power you now have on the problem with that little bit of extra information is you know, sort of it's the whole point of science, right? Like, how do you how do you get there, right? You you lear you get better information on the way, that creates a new experiment, or that creates the need for a new tool. Um, your your information is sort of ratcheting up and you know compounding, and then you eventually sort of burst into this new landscape, and you just you know, it's it's something to behold. We're not quite there yet, but you can sort of see the outlines of this happening at the moment.

SPEAKER_03 49:38

So that I mean that's sort of a a holy grail almost with the the rejuvenation of the aminoque factors without the downside of the pluripotency and even greater rejuvenation effect than than the aminoque factors, right?

SPEAKER_00 49:52

Yeah, yeah. I mean, there's plenty of ways this can still fail, right, as we move forwards. So, you know, we we we do there's boundaries of our assay. So we do a colony formation assay in vitro, we express these genes constitutively, so continuously for six to eight weeks. Um so up till eight weeks, we're not seeing colonies for these single genes. We see them for OSK, we see them for OSKM. Maybe you pushed it another two weeks, you know, they'd start to emerge. I don't know. Maybe we're gonna try this. You know, we're gonna try this gene in an animal, we're just gonna make sure, like, does it, you know, is it safe in the animal? But um certainly if you just compare what's known, like Yamanakopatsis, to what we're finding in the early assays, there's a lot of new stuff that's caused for excitement. Um, so there's certainly these scenarios, right, where some really interesting things could happen down the line.

SPEAKER_03 50:43

So I have to ask a question, and and to someone to someone who's not really hypothesis driven, you know, this may be a completely inappropriate question to ask. But I wonder what the teleology or the biological use for these genes that you've identified. I mean, obviously the the amino acafactors are targeting an embryologic reset, the sort of a natural mechanism. Are the genes you've identified, do you think these are just a subset of the or or maybe a superset of the amino aquafactor genes? Or what are these genes for? Why do we even have them in there? Or is it not appropriate to think of teleology just like we don't need to think of hypothesis? All we care about is results, right?

SPEAKER_00 51:28

So I think there's a healthy level of hypothesis, which is you know, if you just if you just try to join the dots, it might reveal something obvious. You know, like that's there's always a case, right? If you if the dots are in front of you and it's it's a it really looks like a duck, right, without joining the dots, and then you join it's a you know, and that tells you about you know what you what you're dealing with. So just a basic question we ask, which is like where is this gene normally expressed in the human body and when and when is it normally expressed? Um it's expressed in very early development, I believe it's around the time of the natural rejuvenation event, you know, the sort of post-conception, I think it's the first few weeks at least in mice. So that's very provocative, right? The fact that it's you know, in in terms of the timing, it's sort of expressed around that time and then it's sort of shut off for the rest, you know, for the rest of your life. And it's under very strict control. So when we try and over-express this gene, the the biology of the cell is like trying to shut this thing down, it's like methylating the endogenous gene, not the exogenous gene that we're introducing, but it's like absolutely methylating, trying to deregulate you know, downregulate this thing. So that's really interesting. I mean, all transcription factors are very influential proteins, but the fact this one is under such strict control is very interesting. Um, I think there's also this uh association with the germline, so not really expressed across the body, um, but also in the germline. So again, that's that's quite provocative. Um, but we don't know if this is just correlation or causation, right? Like it sort of feels like it could be elegant, right? Like maybe this is some of the biology that's involved, but um, again, we just want to do a comprehensive survey and then like validate everything to the the you know the greatest depth and then see what survives, see what sort of you know turned turned out to be a false positive from our you know screening system. And then I think once you've got really high confidence in the biology you're looking at, that's the best time to try and piece together a story.

SPEAKER_03 53:22

When you when you spoke earlier about mitochondria and the evolutionary evolutionary migration of the majority of the mitochondrial genome to the to the nucleus, it went to chromosome one. Um is that correct? I thought I, yeah. I thought it was one.

SPEAKER_00 53:41

I I I I don't it's it's a long time ago. So I I I I hung up my mitochondrial bits uh around 2018. So it's not that it's not you know it's not that long ago, but um I don't I don't remember that level of detail. Um I'm impressed. I'm impressed that you've got chromosome.

SPEAKER_03 53:59

I never did, I never did mitochondria research at Cambridge, so I shouldn't speak about it. But but whatever gene, it I my recollection is localized to one gene. And even um, you know, Aubrey was talking about moving some of the mitochondrial DNA, which is you know easily damaged in the you know the reactive oxygen in the mind, and move it back onto the chromosome and even move it that way to to get less mitochondrial DNA damage. But I I'm wondering, I just I wonder if any of these genes that you've identified come from the chromosome where the mitochondrial DNA is located, that they could possibly be, you know, at some level mitochondria DNA as well. I hate to keep kicking it back to mitochondria. I was just curious. Yeah.

SPEAKER_00 54:42

So so yeah, so I mean that's a that is an interesting idea. But what what I would say, um what I can say is that these genes we tested, vast majority were transcription factors, and the reason was where did we start? They were yam, they were Yamanaka factors, like that was our definition, and that they were all transcription factors. So inherently, the broader screen, you know, we were like looking for alternatives, so we started with transcription factors. There was some biology that wasn't, you know, it was beyond transcription factors. I don't recall whether there were mitochondrial genes amongst that. Um so I can't I can't confirm or deny right whether this is true or not. The second thing, just to speak to mitochondria, um, I think this is this is a question that deserves um a bit of uh time, which is like what doesn't epigenetic repro like what what are the what are the what are the things that epigenetic reprogramming can't do? And there's there's actually quite a large laundry list, and it's not in my interest to talk about this list, but I think it is necessary to sort of you know appreciate the bigger picture in biology. So there are two genomes, okay? So there's a nuclear genome and there's a mitochondrial genome. So there's basically two operating systems that have to cooperate together. So of course they're both important. Like, of course, it's not like one is in like unimportant and one is important. So epigenetic reprogramming is principally affecting the nucleus. So when we reprogramming, you know, when we reprogram the epigenome, basically taking gene expression in the nucleus back to a young state, that does not reset mutations in the nuclear genome, so we can't undo damage, so like we can't, you know, we can't go back to the blueprinted genome. We don't reset any issues in the mitochondria, at least not at the genome level, so we can't affect mutations that are potentially taking over. Um and yeah, there's many other things, like we don't reverse cells that have already died in the body. So there's plenty of parts of your body where you're losing cells, you're not replenishing them sufficiently, like dopamergic neurons in in your brain, even healthy people lose them. So that's another thing that reprogramming doesn't reach. Autoimmunity. So if you start developing autoimmunity, those antibodies are there, those white blood cells have created those antibodies. You can't, you know, you make them younger, you might actually make it worse, right? They're just producing more autoantibodies. So there's like a whole car crash of stuff that can happen that is potentially irrelevant, m many of it is irreversible. Um, so cell reprogramming relies on the cell still still being alive, not too much damage being in there, the mitochondrial genome still being in like some sort of fit-for-use purpose, you know, fit-for-use state. Um, so it's by no means a perfect therapy. Um, so I'm quite uh you know fond of mitochondria because that's where I came from. I really believe that was a big piece of the picture. And from these transfusion experiments, it continues to sort of show quite dramatic results in certain use cases. So I think both genomes is going to be an increasing theme, like epigenetic reprogramming around the nucleus, like epigenome, um, and then mitochondrial transfusions, right? Like deal with the mitochondrial genome and have more energy-producing organelles. So um, yeah, I don't yeah, getting into this like silver bullet, golden bullet mindset is basically betraying biology, right? Biology is uh there's minimal units in biology, and the fact we've got two genomes is like just the basics of the cell.

Targets For Translation Fibrosis And Hearing

SPEAKER_03 58:10

Yeah, that's that's uh that's such a good point to keep keep that in mind. Um now let's talk about the the good stuff. We could speculate. I mean, that's all good stuff. This is all super super exciting, but um going towards um the whole idea we were talking offline about that we're entering a new era of of potential therapies for you know the age-related diseases that now take up you know 80% of our healthcare dollars by some estimates, these these chronic diseases of aging, we're now seeing the idea that we can approach them with rejuvenation and somehow these rejuvenation strategies will work on possibly all these diseases of aging, you know, assuming there's not like permanent damage like you mentioned on you know nuclear DNA or other, you know, the immune system as well. But but it's a fascinating idea of doing rejuvenation as a strategy for chronic disease of aging uh repair. So with that long-winded introduction, what what what are your targets? You do I mean you guys are away from away from clinical trials uh as well on this. Uh do you what are your what do you think um what what's the low-hanging fruit for for what you've developed here?

SPEAKER_00 59:33

So uh sort of a two-part answer. So we have crystallized these two targets. So one is an expression target, like Yamanacofactors, you overexpress them to rejuvenate our cell. Um so this is a single gene we express, so somewhat like the Yamacofactors, but to our surprise, we also have another target which is an inhibition target. So we can knock down this target uh right now by sRRNA, but we might also be able to do this potentially by small molecule to rejuvenate the cell, and that was very unexpected, right? To be able to inhibit a gene to rejuvenate. The way we found this gene was indirectly when we did this screen of 1500 genes, most of which were transcription factors. Before I said we found 150 genes that were rejuvenative and 40 that were pro-aging. Now, one of those pro-aging genes was very special because it was widely expressed across the whole body. And that's exciting to me because I want eventually the whole body to be rejuvenated. So to inhibit a gene, it needs to be expressed in the first place, right? So it sounds obvious, right? But you need the gene to be expressed if you're gonna get in there and inhibit with a drug. So this special gene was expressed across the body, potentially we could inhibit to do something beneficial across the body. Um, and when we looked at this gene, um in the literature, it had been linked to lifespan, inhibition of this gene had been linked to lifespan, and inhibition had been linked to fibrosis. So the fibrosis was really exciting because it was the first time the data took us to a disease. It's not like here's all these age-linked diseases, which one do we choose? You know, go through all the variables, like sort of do a rain man right on the on the problem space. Um, but this was no, let's do a screen, let's see where the biology leads us, and it led us to fibrosis. So we basically did our own assays because there's a reproducibility crisis, you don't want to be caught up in that. Do your own assays, your own systems that you trust, and we knocked down this gene, uh, we rejuvenated, and then we prevented fibrosis um by using this approach. So these are these are two different sort of you know, two different sides, so overexpressed to rejuvenate, and then knock down to rejuvenate. Um, and clearly that one's tied to fibrosis. Okay, so that's you know, there's a there's a case there to be made, and there's you know, every organ across the body tends to have a fibrotic form in a disease. Like in the liver, it's called mash, uh diet-induced fibrosis, you've got chronic kidney disease, you have fibrosis, you get heart fibrosis. So you can imagine, right, like you could develop a drug even just to a single organ, but for the fibrotic form, and then you could do that for a different organ, like for that fibrotic form, and you sort of got access to most of the body through the fibrotic forms. Um so you know it creates an economy to develop something. Um, on the overexpression side, um, there seems to be a very close relationship between aging and the anatomy of the inner ear. So with hearing loss, the anatomy changes in a very well, not linear way, but in a very predictable way that maps very closely to aging to the extent you can also you can almost use the inner ear as an aging clock. Like it's so it's so tied to age. So um we're actually using that as a use case for expression because the anatomy is tied very closely to aging. And secondly, you can actually get genes to the inner ear by local injection of a virus, so not huge amounts of material that causes an inflammatory reaction, but just a small amount in a site that has partial immune privilege, and we might be able to do something really interesting for a disease that it has huge clinical need, right? You know, you might you might know people that have hearing loss, I certainly do, but it's like a big thing out there that doesn't have a therapeutic, you know, it's just hearing aids. Um, so yeah, we've got hearing loss on that side, fibrosis on the other side. In terms of translation, I think the best description I've got of it like the more you realize about drug development, the larger it look like it looks. It's like a mountain on the distance. You're like, oh, that's a small mountain, and then as you get closer, you're like, oh, that's a bigger mountain, and then you get closer and it's an even bigger mountain. You just keep approaching it, it just gets bigger and bigger, and it and you're still far away from it. And it's it's absolutely astonishing the amount of work that goes into making these drugs real. So I'm I'm not you know, I'm not completely naive to it, but it it every day it continues to amaze me how little I know about drug development. Like the more I know, the less I know, and how much effort it's gonna take to just you know struggle to move these things forward. So um, you know, it's great that we're finding these things in the lab and we're you're sort of crystallizing around indications, but like that's just the beginning, right? Like the amount of risk that needs to be managed, um, progressing these things is enormous. So I think yeah, it's it's almost like a perpetual learning experience, right? As we move forward, it's like, oh, you know, the new chapter has arrived, and you've got to understand this, and it becomes engineering. But I I think it's important to emphasize just how large um and multifactorial problem spec that problem solving space is just for drug development, let alone you know, I've got biology um that's validated. Just just the the journey ahead is very hard. So there could be like a huge amount of promising stuff at the start line, but the finishing line is going to be complete poverty, like just some poor runner that's just collapsing over the finishing line. Like, that's that's almost the you know, it's just it's just the amount of uh trial trials ahead of us is is enormous. So I don't want to put a dampener on it, but I think we should get comfortable with the idea that this is a big challenge, we should gear up for it. We should like get our temperament in the right place, like we're just gonna have to keep going at this, it might take longer than we expect. We should be creative, but we shouldn't like put patients at risk. So I think it's just like sizing up this mountain is an essential activity because otherwise we'll just end up being hopeful and disappointed, and we get jaded, and you know, we just abandon it, and then we come back later because that's really what we were excited about, and you know, it all becomes very disruptive, but like if we size this thing up um appropriately, we can just, you know, do the work that needs to be done. And it's like this expression, right? The shortcut is the long way. You know, it's like just you just need to do the work that's necessary. So that's I guess uh that's something that's more new to me, just the size of the drug development challenge. But it doesn't doesn't put a dampener on dampener on on it for me. But it's something that I think needs to be appreciated like sooner rather than later.

SPEAKER_03 1:06:07

So much great points. So much to unpack there. Let me let me hit a hit on a couple of those. The the hearing loss is uh that's that's great with the with the age-related hearing loss is certainly I mean even rapamycin uh there's animal work showing that rapamycin reverses uh presbycusis which is the age-related hearing loss and and the anatomy that changes is it particularly the cochlear health hair cells or what what is it do you do you recall what it is particularly in the anatomy in the cochlea that changes with the aging I think that's what they were counting with the rapamycin studies it was actually cochlear hair hair cells they could measure those so it may be it may be that I'm not sure yeah yeah so so again I got I guess like we we're taking an agnostic approach so there's a lot of evidence the inner ear hair cells are you know involved right and there's you know there's clearly a loss of these inner ear hair cells that become damaged but there's also changes in the the broader vestibular compartment which is sort of the rest rest of the inner ear so when we when we go into mice for the first time we're actually we're gonna go into as much of that anatomy as possible so the inner ear hair cells and the vestibular compartments we're gonna try and deliver a rejuvenation gene across the board just because we don't want to be presumptuous we don't want to second guess it's just the inner ear hair cells because what if it isn't right what if you need to do both so we're gonna go in broad and then if it works then you can say okay let's narrow it down right let's just just get to the inner ear hair cells and is that enough and then okay that's enough and then we'd be more efficient but again we don't we don't we don't know for sure we don't know for sure and we don't want to second guess so we just try and stay agnostic. So uh again it's sort of unsatisfying to uh hypothesis driven science no there's a beauty to that approach forgive me I was falling back to my old ways of hypothesis driven but I I like your approach and I think I think that's that's a great way to do it and it's obviously uh been successful for you. The the challenges with the animal work versus the in silico AI work it's you know you could you can do you know hundreds of thousands of experiments in a short amount of time with silicon but do you see are you able to bring AI to the animal work? I mean we we talked about the you know the dark labs and that kind of thing do you see ai accelerating that animal work or is it basically now you've hit you know the usual thing and it's going to be years of of work with that so one day one day it's not today un outside of specific use cases but one day we'll have the interventions that we found like in the virtual cell we'll put them into the cells in vitro in a dish we'll put them into the animal some of you know some things will work some things won't work we'll then progress that right we'll progress it down the whole stack of development and we'll be able to train a model across the whole thing and the more the more targets we progress from start to finish it doesn't matter if they're failures or their success although obviously the successes get further right so you have to be quite disciplined but the more we progress the more powerful the model we have that we can train across the whole thing.

Right To Try And Future Guardrails

SPEAKER_00 1:09:20

So there will be a point it's gonna it's gonna require a lot of investment um against a a landscape where you know it's hard to raise money for drug development like let alone make the experiment broader right to generate more data but eventually there'll be a model where at the very early earlier stages based on the data beforehand you can predict whether that's intervention is going to be a clinical trial success. Obviously there's going to be error bars right and there's going to be a certain confidence but there will be a point and these models are being trained right now there will be a point where at the earliest stage you can predict if this is going to be a clinical trial success in humans. So the use case that has been built is for liver toxicity. So um basically there's a lot of data for molecules that have worked and haven't worked between the earlier stages and liver toxicity in in humans right there's sort of a critical mass and a model's being trained across that and so you can actually at the very earliest stages um assay for liver toxicity like basically in a in a simulation and avoid that issue which is one of the big causes of clinical trial failure right that something gets liver toxicity so like bioH right their their intervention it was a liver toxicity issue. So this is you know that's that's one use case but eventually there'll be a model that crosses the whole the whole thing and you'll be able to say is this going to be a gene that can affect lifespan obviously health span and lifespan right that's you know I that's my that's my wish right health span and lifespan together but you'll be able to sort of you know in the future I don't know how long this will take how much investment but you'll be able to say like will this affect health man health span and lifespan in human beings with certain amount of confidence. That's pretty cool.

SPEAKER_03 1:11:02

Yeah yeah it's so exciting I want to be respectful of your time just maybe a couple more a couple more points if you if you can that we talked offline about the FDA and you know the challenges of uh a clinical trial the expense and and you know the theme of this this conversation a lot of it has been don't use hypothesis well the to step away from hypothesis driven ideas and be open to what the data shows us and you know with AI that's possible and all well the FDA is almost the antithesis of this they are asking for you know a definite hypothesis and and so it really limits your choices and um it it's really you know it's expensive it takes time um so what do you think about the role of citizen scientists about um expanding you know testing with drugs obviously it'd all be done within a l legal framework of course you know but um the idea to you know once safety is established at least open up to you know many possible many possible chronic diseases great question so uh the nature of what we're doing is very expensive so this sort of target discovery for cell rejuvenation genes is some of the most expensive basic science you can imagine like you we we eat single cell mixed reagents for breakfast lunch and dinner and these these are like some of the most expensive experiments like you could do in an academic lab.

SPEAKER_00 1:12:34

So you know the the burn rate right is just sort of makes my eyes water sometimes and where do you where do you find the money to do those types of experiments? It needs to come from meaningful capital pools. So you know we have the initial check and shift was relatively small but then eventually you you have to enter VC territory and the the critical mass of capital is like biotech VC right and they they want this to live in biotech right that's just that that's where the capital is and that's you have to develop something there. And then you get to the situation where like there's these timelines and there's all the you know there's regulatory checkpoints and there's sub sequential financings you have to get through and you have to sort of please a lot of people it becomes this sort of this thing that you have to manage right that's very different from the basic science. Like do you do you continue to plug away right like just struggling on just putting up with it or do you like get out of that system right and try and do something on a more aggressive timeline and sort of manage the risk yourself just try and create something in a different system. As a developer of drugs with backing from biopharma my best answer at the moment is we have to build it in biopharma because if we build it in biopharma it can cut across to the other side. If we decide to do something over here instead of biopharma um it's very hard to cut across back to biopharma and and I I sort of see biopharma as a distribution system. If we build it here it's going to go everywhere very quickly if we build it over here it's gonna be in the long run it's gonna be harder to distribute and so I think I think we have to build it in biopharma like that's not optional. I don't think we can abandon that I think there's perhaps there's processes that are slower than they need to be um but also there's a lot of really great people there and checks and balances that are there for very good reasons. The idea that as a startup company I'm going to reinvent that whole system on top of what I'm doing which is this enormous you know mountain that needs to be climbed I I think you know we're finally spread already let alone reinventing the whole drug development system. So I think we have to build something there but I also recognize that there's there's people that they're in a lot of pain today right or they're running out of time like they've been given two years or something like that. They they can't wait for this thing to take 10 years, 20 years, right, in in its fully flourishing form. So I think I think there should be a button somebody that can press before cryo preservation there should be a button someone can press which is like I'm out of options and there's this amazing technology today there's sort of the um I guess surveillance of AI like when you're doing new things right you can be be more careful about what's going on and shut it down. So you've got all this amazing technology um and I'm not able to press a button and use the best that's available ahead of you know so I I think there's like compassionate use there's there's sort of different versions of this but I think that's I think that's something that should be formalised. I think just saying you know it is the way it is right you've got two years there's nothing we can do you should go to a hospice enjoy the time that's left I sort of feel like you know I think there's I think there should be like a courageous hospice. I'm not saying every you know I don't I don't want to I don't want to frame people yeah because that's you know that's that's a really difficult situation there's a lot of courage in there but I think there should be a sort of a proactive hospice where people are like well I'm gonna try something right and I'm I'm out of options and I'm willing to try something when otherwise it wouldn't be try you know wouldn't be tried.

When He Nearly Quit

SPEAKER_03 1:16:12

So I I think it'd be very helpful to formalize something there um and you know sort of build that into the other system and and you know I think that's gonna take a lot of people right to sort of create that that's not you know that's not something I could do on the as a side project right you know that's uh that's an undertaking in its own right but I think there's appetite out there you know that's why I appreciate uh sort of Tom Benson and what he's done with Mighty Nortes right it's like a proactive approach to your situation instead of you know just making do um you know making do with your lot yeah it and it's we talked a little bit offline about that yeah we're I'm working with a group that in in the United States many of the states are starting to do these right so-called right to try laws and originally it was for people who only had six months to live you know compassionate use but now at least one state Montana has dropped that so any you know adult of sound mind uh can can take drugs right now they still have to have passed phase one clinical trials with the FDA but even that may be rolled back because you know as as adults we should be able to you know make choices about what we do with our bodies at least well I don't want to get political or anything but but it there's a there's a growing uh sense of that in the US and there are there are places now that that has been codified the regulations allow it and so I think we're gonna see more and more of that and if the regulatory framework is changing so if we now just need kind of a good scientific framework so we get all the data we don't lose the data and just have a bunch of people in the wild west trying it but but I I I think I think like a version of this I don't know I don't know exactly what it's gonna look like but if there was some AI or you know narrow medical AI looking over your shoulder like plugged into you whilst you're trying these things you know like trying to trying to make this as safe as it can be I think that would be very helpful some like massive scrutiny device that has you know it's machine learning is built into it and helping people navigate the front lines of medicine like doing like being the n equals one for the first time don't hit this landmine right they think this thing can give you an early warning system ahead of any human seeing that like I feel like that's a very promising direction like yeah let's make the most of today and these tools that are becoming available to allow people to explore well previously it was huge huge amount of sort of treacherous risk and yeah it'd be really nice to I don't know exactly how this is going to get formalized but it just feels like something that should be built yeah yeah I mean it it's definitely coming it's already happening so uh hopefully that will that will come sooner rather than later and this this has been great anything we didn't cover Daniel today that you want to end end with or uh this this has been such a great conversation um I think I think just there was one question I put on my top five questions I'd like to be asked. Okay.

SPEAKER_00 1:19:18

Um and it's when did you get closest to giving up so I sort of described this um yeah this this experience of seeing a mountain really like covering a lot of distance but the mountain's still the same size and then you're like oh it's bigger and then you keep going it's like oh it's bigger so I think a perfectly rational response is when is it gonna get bigger? Like for like I'm playing the long game here but if if if if winning is on the other side of death right it's not really a game you know it's like I'm just you know yeah so like like when when did I get closest to giving up so there's there's a few examples of this where um I think I'd been in cell culture for like three or four months and I should have only been there two months because you know um that was the duration of the experiment but like about two days from the end all my cells got contaminated right this is something that happens in cell culture to reset the experiment it's like two months down the drain. So like you set it up again and then you're like really you know you're overly policing the area you're being quite mean to people because you suspect everybody was you could just see like this you just your characters getting you know you're getting in a bad place right and it's not you know you're sort of miserable and but you know I I sort of you know it was like if this result doesn't come in I remember there was this instrument called the Pyre sequencer like I knew what the shape of the the graph would look like if we'd reduced mutations in the mitochondrial DNA. I should my heart was beating so it's like four months of work right it was it was it all for nothing and it came in and it's absolute like euphoria right like you you sort of struggle like you haven't got any melanin right because you've just been in like a bunker in cell culture and like and like I was it was like if this thing doesn't come in that's it right this is not you know this this is crazy like this situation is crazy but then it came in and it was absolutely worth it and many times I've been in these situations where you know you see you know your friends like moving on in life there's things you want in life that aren't just solve aging right there's like other things you see people moving on and like meanwhile you're just stuck in cell culture and they're like bringing you like oh we're having a great time where are you and you're like oh I'm with my cells and they're like not again but yeah but but each time I've I I've had a number of these right where like if it doesn't work I I you know I walk away and but it's always worked but it's it's a it's a you know you worry you worry that it's only going to fail later like the damage is only going to be larger because there's you'd never know for sure right it could be like this initial hypothesis right like I believed I'd solved aging and I was waiting for this signal but that wasn't actually solving aging so you always worry that now you've wasted more time you have to go back to square one um but there is a point there is a point where you've got sufficient breadth on what you're doing where there's like five shots right and one of those is going to work and so like we're now in a place right where there's enough shots on goal that we've got a critical mass that one thing could keep moving us forwards and then we create diversity around that and we repeat um so like I just I'm very grateful I'm out of that hit or miss zone right where you're very much just fodder to probability right like you'll you'll be very lucky to make it out of that situation where you're like you know very narrow as a company um you've got one hypothesis like that sort of stuff so like we're in a much better place now.

SPEAKER_03 1:22:43

That's a that's a great life message it's a great place to end it I I'm I'm sorry to to dive too much in the technical stuff when there's so many good messages like that too it's the professor and me I think wanted to pick your brain a little bit about that but this has been a wonderful wonderful conversation thanks so much Daniel for for the for the great work you're doing and for spending time with us and uh we want to have you back on the podcast soon.

SPEAKER_00 1:23:11

You're very welcome Robert and uh thanks everyone for listening.

Subscribe Reviews Sponsors Disclaimer

SPEAKER_03 1:23:16

If you're enjoying this program please hit that subscribe button or even better leave a review. Your support makes it possible for us to create the quality programming that we're continually striving for. Half of all heart attacks strike without warning your best early warning a CT calcium scan. We've made it even better by adding arterial age and liver fat quantification all from the same scan. Now available nationwide without a doctor's prescription use the code LUFKIN20 for 20% off. This episode is brought to you by ElNutra, maker of the Prolonged fasting mimicking diet. Use the link in the show notes for 20% off. One of the most common questions I get asked are which blood tests I check for myself on a regular basis. Well I check up to 50 biomarkers but the secret is that I'm able to do everything from the convenience of my home and I collect a sample with this new breakthrough device. So no more trays full of tubes of blood, needles or even finger sticks and you don't need to be a doctor to do it. Then I just mail it in for the results. CyFox Health makes it possible use the link below to get 20% off can I start is it's already recorded.

SPEAKER_02 1:24:39

Oh sorry this is for general information and educational purposes only and it's not intended to constitute or substitute for medical advice or counseling the practice of medicine or the provision of healthcare diagnostics or treatment or the creation of a precision patient or clinical relationship. The use of this information is that the own users risk if you find this to be on the value please hit that like button to subscribe to support the work that we do on this channel. And we take the your suggestions and advice very seriously so please let us know what you'd like to see on this channel.

SPEAKER_03 1:25:21

Thanks for watching and we hope to see you next time you save the recording