借助AI驱动药物发现：Isomorphic Labs总裁Max Jaderberg

Much of the pharmaceutical industry relies on a fairly empirical process for creating drugs. Problems arise during clinical trials because we don’t see them until we start giving the drugs to patients and discover unexpected toxicity. Therefore, it is important to gain as much understanding as possible before the drugs reach patients. How can we transfer this entire drug development process to a computer? How can we go beyond the real world? I would be very disappointed if, in five years, the predictive understanding of disease biology had not taken a huge step forward. Do you have any guesses as to when, for example, 80% of human cancers will become curable?

1. Max Jaderberg Introduction

Today on No Priors we have Max Jaderberg as our guest. Max is the president and co-founder of Isomorphic Labs. Isomorphic is backed by Alphabet and other investors, and its goal is to defeat all diseases. We’ll talk about the role of artificial intelligence in the future of biopharmaceuticals, human health, and longevity. Max has a background in machine learning and a PhD specializing in machine vision. He eventually sold his company to Google DeepMind, made a career there, and is now the president of Isomorphic. Max, thank you so much for joining us today on No Priors.

2. Isomorphic Origins

I am very pleased to be here. Thank you.

Isomorphic, in my opinion, is one of the most interesting companies working in technology right now. Alphabet is a major shareholder and sponsor, and Demis, who built a lot of models at DeepMind and elsewhere—you sold your company to him, I think—really helped with the founding and launch, and now runs it. You’re tackling one of humanity’s greatest challenges: human health. Could you tell us a little about how you started working on this company and what drove its creation?

I was part of the founding team of Isomorphic Labs. Before that, I worked at DeepMind for a long time. Personally, I was working on early generative models, language models, deep reinforcement learning, and big challenges like games. That was my passion: AlphaGo and AlphaStar.

Eventually, we got to the point where we were taking on what people called impossible tasks. This was probably in 2018 or 2019. We were asking whether AI could win video games and things like that. At the time, we were still trying to prove to the world that AI was capable of anything at all. We took on these impossible tasks, and I saw firsthand how to solve them systematically.

For me, there came a moment when I realized: We haven’t reached general-purpose AI yet, but we clearly have this incredible set of technologies in deep learning and reinforcement learning. What problems should we really focus on? I was thinking about that. Demis was thinking about it at the same time. He saw the results of AlphaFold, the first signs that it was starting to work, and for us it was the first proof that we could create a deep learning model that generalized across a significant space in biology.

That proof of concept convinced us that the time was right for this big mission: changing the way we approach healthcare and trying to defeat all diseases. That would require a serious effort, our own company and culture, specialists from different industries, and the ability to move very quickly. That’s how Isomorphic Labs was born, around 2021.

Oh, great. Your journey began primarily with deep learning and AI modeling. I think you sold your company to DeepMind in 2014, right?

Yes, that’s true. Quite early.

You were doing this before most people realized how important it was. Can you tell me more about your journey and how you came to this?

Of course. What interested me at the very beginning? I’m an engineer by education. I ended up in graduate school in computer vision almost by accident, and it happened at a turning point. I think I started studying before AlexNet came along.

AlexNet came out in 2012, right?

Yes. Then I saw it develop, and I thought, “Wow, this is incredible.” I remember talking to Yann LeCun about convolutional neural networks, or ConvNets. I thought we should incorporate them into the core computer vision research we were doing at Oxford.

Eventually, I developed all these methods. We had the winning network in ImageNet 2014, with the best object recognition and text detection, all in that vein. I started a company in computer vision as a service and was trying to raise funding. At the same time, Demis and DeepMind had just sold to Google.

While trying to raise money quickly, I approached Demis for angel investment, and he said, “Why don’t you just join DeepMind and build general-purpose AI together?”

What is general-purpose AI?

That began my journey at DeepMind and everything that came with it.

3. Drug Discovery Stagnation

It’s amazing. Now you’re focused on a vertically integrated company that does everything from models to biology. The idea is to find medicines that will help people with various aspects of health.

If you look at the pharmaceutical industry, it’s striking that the number of new drugs approved per year remains almost constant. I’ve even made graphs. And that’s despite a veritable explosion in knowledge. The number of scientific publications has grown exponentially; the number of articles has increased 10 to 20 times over this period.

The number of new modalities and types of drugs has also increased. Now there are antibody-based drugs, small molecules, proteins, peptides—all these new treatments. The cost has increased a lot, too, hasn’t it? It went from several hundred million dollars for a drug to $1.5 billion. Development time increased from 8 or 9 years to 15 years. So all the indicators have worsened, while the number of new products has stayed about the same. The knowledge base exploded, and the number of available types of molecules increased significantly. Why has so little changed despite such progress in other areas?

It’s incredible that developing new drugs is becoming increasingly difficult, expensive, and time-consuming. I think there are a lot of reasons for that. What I think about a lot is that the way we create medicines, both in industry and as a species, has always been a very empirical process.

In the very beginning, about a hundred years ago, medicines appeared by chance. We happened to stumble upon something that people understood had a certain physiological effect, and we started using it. The process became a little more complicated, more like empirical trial and error. We guessed at a lot of drugs. For example, for a long time we had no idea how aspirin worked. We didn’t know how paracetamol worked.

The way we created drugs, and the way much of the industry still creates them, is very empirical. You have a hypothesis about how a disease might develop and what a good molecular design might be, but you need to create it, test it in real-world conditions, and do that iteratively.

It’s slow, and it’s very difficult to get results from this process. Biological space is very large and complex. A cell is a big, complex, chaotic system. It’s extremely multidimensional. Molecules—whether small molecules, antibodies, or peptides—are also very multidimensional objects, with combinatorially huge design spaces. The only tools available to us are essentially trial-and-error screening processes.

For me, that limits the complexity of the solutions we can find with the limited time and money available for this empirical work. Over time, we’ve found the “low-hanging fruit” of drugs through our empirical development process. We find the easiest things, but it becomes increasingly difficult to increase complexity and find things beyond the reach of that process.

This is interesting. But don’t you think that, if you look at any other industry, the age of the industry often affects its rate of change? The older the industry, the slower it moves. Everything becomes more bureaucratic, and more regulatory barriers appear.

It seems like the last time a biopharmaceutical company worth over $50 billion emerged, other than Moderna—and I think Moderna’s success was almost a fluke because of COVID—was in the ’80s. I think it was Regeneron.

Yes. Regeneron, Vertex, and Amgen were all from that era. Amgen was in the early ’80s.

How much do you think this comes down to regulatory problems, bureaucratic organizations, or just natural aging? It happens everywhere. It was like that with the automotive industry before Tesla came along.

That’s right. We’ve talked about the scientific aspect of this issue. I think there’s also a big cultural aspect to how industries continue to innovate. How do you keep moving really, really fast? And, related to the scientific and technical part, what mechanisms provide economies of scale in innovation and research and development?

If you take on a specific goal to create a drug and give it your all, what do you take away from that experience to work on the next drug faster, and then even faster? If you can accumulate this knowledge over time, you can gain the benefits of scale and build something truly amazing as a company.

A lot of the methods and approaches to drug development in the industry treat each project as a stand-alone, unique scientific experiment. You learn so much about that project, but not much of it can be transferred to the next one. Of course, there’s a lot of knowledge in the different disease fields.

Yes.

And this is where, in my opinion, everything can change significantly with AI and machine learning. If you have a drug development platform that allows you to truly absorb knowledge, understand the underlying biology of diseases, design effective molecules, and craft these drugs, then you can expect to gradually accumulate advantages with each new project.

One of the most impressive things to emerge from deep learning and the era of foundation models we’re in is the ability to generalize at scale. It seems that in biology, too, if you have enough scale and a sufficiently thoughtful approach, you can generalize some of these problems that used to be purely individual.

So what approach do you use?

That’s a big shift we’re making at Isomorphic, because we’re not the first to try to change how we develop and find drugs with AI.

4. Hiring at Isomorphic

One of the most exciting things about Isomorphic is how diverse the team is, working together toward a common goal despite often having very different professional backgrounds. Can you tell us more about the kind of people you hire, what their profiles are, and who you’re looking for right now?

This is an interesting field, because the combination of advanced AI with drug development, chemistry, and biology hasn’t existed for very long. So you can’t just go out into the market and say, “Hey, I want to hire the world’s AI experts to develop drugs,” because they don’t actually exist. And if they do, they don’t use our models; they apply other methods.

That’s why we try to recruit the best experts in basic AI and deep learning. They may not have a background in life sciences, biology, or chemistry, but they’re genuinely eager to see how their work in deep learning can bring real benefits to humanity. We hire AI professionals and biology and chemistry specialists who may not have had much experience with AI before.

Then we work hard to build an internal culture where everyone communicates and learns from each other. As a result, we have interdisciplinary experts and scholars, which is very interesting to watch.

What do you think about the ratio of these specialists? I started a company many years ago called Color and was its CEO for the first few years. My co-founder led it over 10 years ago. It was quite a while ago; we were using machine learning and data science in genomics and building genomic pipelines. We found that you could have, say, five machine learning experts and one scientist from a traditional field, and that was a pretty good ratio.

Mm.

How is it now? How do you determine the optimal number of people from each specialty in a group working on a problem?

For some of these key AI projects—super projects, AI modeling projects—I don’t think it’s really that different. You have the core modelers, like engineers and AI scientists, and then you just need key industry experts: people who really understand the application and the nuances of the data, like what the metrics mean.

They make sure we’re not chasing false metrics, and that these are real things that will lead to the creation of a cure. Those ratios of one to five, I think, pretty accurately reflect what our teams look like.

Are these specific roles that you’re actively hiring for right now, or are you looking for employees across the board?

5. Bringing Drug Discovery Onto Computers

We want the best AI researchers to come to Isomorphic and help us advance this mission. There’s literally so much work to be done at MGM on basic simulation models like AlphaFold plus plus plus, and also on basic reasoning models—for example, taking all the cutting-edge open-source models and adapting them to our field. I think this is a very important area for our hiring.

What components do you cover? What are the main areas of focus in your approach to new drugs? Do you have your own labs? Do you use robots for pipetting? What exactly is your approach?

If you go to our core AI team, our main focus and what we’re really trying to achieve is moving all the work of drug discovery, drug design, and understanding disease biology onto the computer.

Mm-hmm.

How can we go beyond the real world?

Mm-hmm.

A lot of people are developing ways to interact with the real world very efficiently, so you can do a lot of iterations in reality. This is great and complements us, but fundamentally, the main bottleneck in time and cost is synthesizing real atoms.

Mm-hmm.

And then testing them in real-world analyses and experiments. As we’ve seen in almost every other field of engineering, moving to computer-aided design systems has changed what we can design and create. We want to do the same in this area.

When we think about developing a core AI engine for drug design, we focus on achieving experimental accuracy.

Mm-hmm.

That’s what we want for our simulation models, our models of the world. Take AlphaFold as an example. AlphaFold predicts the structure of proteins and how they interact with each other. Typically, obtaining a structure through X-ray crystallography takes months or years. If you can do it with a simulation model like AlphaFold, you don’t need to go into the real world. You save many months.

Creating drugs is a combination of many very complex scientific and engineering problems. If you can build predictive models instead of doing real experiments, you save a ton of time. Internally, we have a line on all our leaderboards that denotes experimental accuracy. The goal of all these “superpowers” we’re trying to create is to achieve a level of experimental precision.

At that point, you no longer need to go out into the real world. You can do a lot more design and iteration on the computer, so that when you finally get to the real world, create molecules, and test hypotheses, the probability of success is much higher than with traditional methods. Then you don’t have to try as many things in the real world. Everything becomes faster, and problem-solving becomes easier.

When I look at drug development, I think it takes about 15 years now and costs $1.5 billion. In my opinion, work with small molecules—in vitro, tissue cultures, and experiments on mice—takes, say, two to three years and costs about $50 million. That’s the order of magnitude. Then another $1.45 billion and about 12 years goes into clinical trials, the different phases, safety testing. How much of this do you think can actually be transferred to a computer?

There’s a discovery phase where you go from an idea to the point where you think you have a good molecule that can be used in patients. Before that, preclinical studies on animals need to be conducted to make sure it’s safe and works as intended. This takes another year. Then you move on to clinical trials. There’s a first phase, a second phase, a third phase, and the load increases.

When you look at the graph and the failure rate, I think that’s what drives the average cost up so rapidly. Why is everything so expensive? Because so many attempts fail.

Mm.

The real failures happen in the clinic.

Mm-hmm.

They fail in the clinic because we could never see the problems until it got to the point where we give these drugs to patients and say, “Oh no, there’s a toxicity we didn’t account for.” Or we have a molecule that acts on the target we designed it for, but also affects something else, with a very negative effect.

We may be working toward a cure for a disease, as we thought. It may even cure it, but the drug itself has a very negative side effect.

Mm-hmm.

People only learn about these problems when they start administering the drug to real patients. So the big opportunity is to understand much more before you get to patients.

Mm-hmm.

If you better understand the biology of the disease and how cells work, you may be able to understand exactly how what you’re trying to influence is related to the disease.

Modeling all these other components, not just which molecule will bind to what.

That’s right. We model all these components.

6. Isomorphic’s Datasets

How do you get the data to model these things? Where do these datasets come from?

We have an incredibly comprehensive data strategy. Data is scattered everywhere, but I think the main thing we’ve learned in the last five years or so is that there are a lot of opportunities to generate this data yourself.

Of course, there are large data repositories. Sometimes they’re very useful for deep learning. Often these repositories, whether public, pharmaceutical, or corporate, are a byproduct of years of passive data collection during drug development. So they can be very biased, which often doesn’t help if you’re trying to find a one-size-fits-all solution for drug development.

If you’re really interested in generalizing these models, you may need a different data distribution. So we generate this data ourselves. We’ve seen some really amazing data scaling laws as a result of this.

7. Rethinking Endpoints

How do you measure whether some of these things will have side effects? There’s an old saying—I have a PhD in biology and have worked in various fields, including oncology and neuroscience. In oncology, we’ve cured cancer in mice hundreds of times, but these molecules didn’t work in humans. So a lot of the data you actually need has to come from human clinical trials.

How do you approach generating data that is truly informative? After all, if you use animal models, the resulting data often doesn’t correspond to human biology.

Well, actually, yes. It seems like you’ve experienced all of this firsthand. Very painful.

When it comes to toxicity, for example, the gold standard regulators require is proving that your molecules aren’t toxic to rats and other animals. But even when drugs are approved and tested on rats, they often turn out to be toxic to humans. Why is that? A rat isn’t really a person. It’s not a very predictive model.

I think it depends on the study, but rats are about 30% predictive of clinical toxicity in humans. If one of my AI scientists or engineers said to me, “We have an incredible model; it’s 30% accurate,” you’d say, “Okay, maybe we should do a little better.” There’s a lot of opportunity to rethink what the predictive indicators should be for human clinical toxicology.

Let’s forget about animals. They’ve been substitutes for the last 50 to 70 years, but perhaps we can do better. We can combine in vitro metrics that are easy to measure in the lab with AI-based predictive models. There are also amazing new experimental technologies emerging in organoids and organs on chips, such as synthetic human tissues that mimic human organs and create their microenvironments. One could imagine combining all of this to model human clinical toxicology directly.

There isn’t much data available right now because only a limited number of clinical trials have been conducted. But I think we now have very good methods.

8. External vs. Internal Dependences

Regarding these approaches or techniques, do you usually look for partners, or do you develop them yourself? I’m curious how you operate as a company, considering how much can be done. I’m curious how much you want to focus on computer science as well as biology, where CROs—contract research organizations—are often used. Unfortunately, the quality of their work often varies. How much do you rely on external parties, and how much on internal resources?

We build our drug design engine and all of our AI models ourselves, and we see a big opportunity in creating and owning our own data. But biology and experimental platforms are complex, multifaceted fields, and there are incredible people working on fantastic platforms. So we work with the whole world: ourselves, contract research organizations, and other innovative companies to create the data that feeds our models.

9. Automating Biologic Labs

This is cool. Listen, Anthropic recently announced the creation of a wet lab in San Francisco. Reading between the lines, based only on external publications, it seems like they’re trying to build some kind of robotic system or closed loop. When you look at AI models, as you know better than I do, you need closed systems to iterate quickly. If you get quick feedback, like when writing code that can be tested instantly, you can quickly check whether the model is working well or not, right?

In biology, one problem is that you do an experiment, wait three weeks, and then get the results, which still have to be interpreted. That’s a big cycle. It slows progress, the ability to iterate on data, and the creation of robotic platforms that could speed things up. How do you approach this problem? Do you think this is part of the solution you’ll all strive for?

It’s been very interesting to see how much better automated labs and robotic solutions for experimental labs are getting. We’ve been working with these automated labs for a long time. Everyone should be doing this, and we should continue to reduce the iteration time with the real world.

But fundamentally, I want to go beyond the real world. Can we really do that? You’re still going to run into the limits of physics in the end. How long does it take to synthesize a molecule? At some point, you just have to wait for the reaction to happen. How long does it take to culture a cell? At some point, you have to wait those days for the cell to reach the right state before you can test anything on it.

So yes, I think we can do a lot of work, and we should. As a species, we’re also working on improving automation and the ways we test.

Sounds like a transfer to an in silico environment. You want to bring this into the real world.

We’re very keen to implement in silico technology on computers to reduce the time to real-world testing. This will allow us to do much more predictive work virtually, and we’ll only turn to the real world for validation, not for training or repeated iterations.

We still need to go into the real world for validation because we’re putting forward new scientific hypotheses and ultimately want to test them in humans. We need to make sure they work and are safe. Going into the real world for validation is necessary, but the question is how much time spent on real experiments and iterations can be reduced.

10. Internal Development vs. Partnerships

How much of your approach focuses on partnering with existing pharmaceutical companies, and how much on developing your own drugs?

We’re developing an AI-based drug design system ourselves, and we don’t sell this model or grant a license to use it. We don’t sell access to it in that way. We use it internally to create medicines.

Mm-hm.

We do this in 2 ways. We have drug discovery programs that we fully control, fund, and develop ourselves—our internal project portfolio. We also have programs that we work on together with great pharmaceutical and biotechnology companies, where we act as partners.

11. Areas of Human Health Focus

That’s exciting because these companies and their teams often have amazing ideas and expertise in specific areas of medicine. They come to us with the most complex tasks, which they may have worked on for 10 to 20 years without making much progress. Then, thanks to our drug design engine, we make tremendous progress together.

What areas of medicine or human health do you focus on?

Internally, we focus primarily on oncology, immunology, and inflammation.

Mm-hm.

The most interesting thing is that our drug design engine and model are universal. They’re not limited to certain diseases, types of targets, or therapeutic modalities. They work with different treatment methods.

Starting with cancer, is it autoimmune diseases, or something else?

Yes. Immunology and inflammatory processes are often closely related, so we work in this area. When we tackle this internally, whether it’s oncology, cancer, or inflammation, I think there’s a great opportunity to tackle some of these root problems—things that could have a positive impact on many different patients and follow-up indications if we find a solution.

Can you give an example?

I mean, if you think about cancer—

Mm-hm.

—it’s a very complex disease with a long tail, and it manifests itself in patients in many different ways.

Mm-hm.

But we also know there are some basic driving mechanisms that may be common to many different things. We saw this amazing example earlier this year: a KRAS inhibitor received a standing ovation at a major cancer conference for dramatically improving the lives of patients with late-stage pancreatic cancer. That was from Revolution Medicines, and it’s a great example of how we’re tackling certain root causes.

I understand.

Yes, RAS has been in the cancer literature for 20, 30 years. These problems are incredibly difficult from a drug development and design perspective, given how precisely you need to modulate things. And, of course, in cancer, everything mutates very quickly. You think you’ve found a solution, but clinically it doesn’t work because the disease mutates around it.

12. Multidrug Approaches

This is an interesting point because this was also the case with many different drugs and areas of healthcare, like HIV, right? You end up with a cocktail of drugs, and that’s what allows you to suppress AIDS or HIV. In cancer, many drugs were developed as single drugs, not as cocktails.

Mm-hm.

As you mentioned, there’s often a spectrum of mutations within a tumor that need to be suppressed simultaneously if you really want to cure it. Otherwise, it will evolve given the existing distribution of mutations in the cells.

Are you thinking about more complex cocktail or multi-drug approaches?

So why aren’t we prescribing more drug cocktails? The problem is that these molecules often work, but they have some toxicity or an effect on the liver. If you take several of these drugs at the same time, the interactions can be very harmful to patients. Because we’re unable to create perfect molecules and drugs, we can’t safely combine them for patients, which limits their application.

Yes. I think there are only a limited number of patients who can participate in a clinical trial. The combinatorics of the process also become more complicated when it comes to clinical trials, especially when there are a lot of risks and uncertainties about how these things will come together.

This is where we see an opportunity. If you can design these molecules much more precisely so that they’re safe, don’t affect the liver, and don’t interact with each other in an undesirable way, then you can create quite complex combinations that modulate not just one target at a time, but a whole vector of targets across complex biological pathways at the same time. That can have a better impact on the disease.

There’s a very interesting video on YouTube about Paul Janssen, who founded Janssen Pharmaceuticals. The company was eventually acquired by Johnson & Johnson, and he was one of the great pioneers of pharmaceutical drug development. In my opinion, Janssen was one of the most productive pharmaceutical companies in terms of the number of drugs it released. In the video, Paul Janssen strongly argues that the main limitation in drug development is an overemphasis on safety.

He says regulators are holding back progress. He noted how, during World War II, Winston Churchill tried to find a cure for soldiers with venereal diseases in the field. They rediscovered antibiotics: penicillin had been discovered by Fleming in the 1920s, then shelved and forgotten. Within nine months, they rediscovered and commercialized penicillin from scratch specifically to treat this disease. This happened because all regulatory obstacles were removed. That was the example Paul Janssen pointed to.

13. Risk Appetite Variation

It can be argued that the COVID vaccine had a similar development speed. How big do you think the overemphasis on safety is? As Paul Janssen pointed out, it’s a question of risk and reward. You have to make compromises, and you can’t categorically state that the main thing is safety. It’s also a matter of profit, and in some cases you should take risks. What do you think about this?

Honestly, the most important thing is to always keep patient safety in mind. But then you need to understand what the risks are and the patients’ willingness to take them, which may vary depending on the indication. COVID is a good example: patients’ willingness to take risks changed dramatically because of the prevalence and danger of the disease. In my opinion, this changed demand and forced us to act somewhat differently. This is noticeable even in different fields of medicine.

If patients have late-stage cancer and few treatment options, their risk tolerance is completely different compared with cases where you’re trying to treat chronic, low-grade inflammation. I think it’s very important to focus on patient safety, while adapting to patients’ needs. Ultimately, we want patients to live healthier lives and receive treatment as quickly as possible.

We communicate a lot with regulators like the FDA here and around the world. What they have in common is that they’re all sincerely committed to providing patients with the best medicines and solutions. Over time, I think we’ll see differences in the success rates and safety profiles of drugs developed with a deeper understanding of their mechanisms of action. We’ll be able to look for better design solutions than were possible with an empirical approach. I hope to prove over time that this produces statistically significant differences. That will change our approach to regulation and the way patients get their medicines.

14. Disease Treatment Outlook

When do you think you’ll start the first clinical trials?

I’d really like to give you the exact date. This is truly advanced and extremely complex science. We’re working on some very interesting drugs, and everything is going well. We’re making efforts to deliver them to patients as quickly as possible. We’re expanding our clinical development team in preparation for this.

I know these are always impossible questions. Do you have a prediction for when, for example, 80% of human cancers will become curable? Or some other metric that shows why this is so important to the world?

I would be very disappointed if, in five years, we didn’t have the basic tools—if our predictive understanding of disease biology hadn’t taken a huge step forward. Our ability to design molecules that are significantly more precise and less toxic hasn’t made such a huge leap, so drug development and discovery still take years. Over time, this will decrease, but right now it takes years. We need to have the tools to start tackling large parts of disease in a very different way.

15. Human Longevity and Drug Development

If cancer and heart disease were removed from society, statistics seem to indicate that the average person would live five to seven years longer. So the real impact on longevity from the two biggest killers is, in some ways, quite large and, in other ways, quite small.

A number of companies have tried to focus on aging or longevity as a more general process that’s obviously genetically programmed. If you turn off certain genes in C. elegans, Drosophila, or mice, they live 20–30% longer. In the case of C. elegans, little worms, they can live three times longer if you turn off certain genes in certain pathways. I was actually working on these pathways. This was part of my dissertation, surprisingly. It was very interesting.

How do you view this in the context of human lifespan? Will we eventually reach a point where people live three or five times longer? What do you think about longevity in the context of drug development and all these tools that will now exist?

This is interesting because there are major diseases that affect people, and if we could start to influence them, it would specifically affect life expectancy. Beyond cancer and heart disease, other chronic ailments of aging are starting to accumulate, and their root cause is literally aging. In other words, the root cause is that these systems are allowed to break down over time.

They break down and combine in such a way that, even if people live to the ripe old age of 80 or 90, they may spend the last 10 or even 20 years of their lives not as well as they could have. Your bones start to break down, your muscles get weaker, you can fall, and suddenly you’re not as mobile as you’d like. This affects your mental health, and suddenly your quality of life drops dramatically. You can still live another 10 years.

In the aging research community, this is referred to as lifespan versus healthspan. We think about this a lot, and in all models of aging, if you influence aging and delay it, healthy life expectancy also increases. If C. elegans lives three times longer, it actually stays a healthy adult three times longer, then eventually declines. You’ve extended both at the same time.

I’m curious what you think about life extension and things like that.

It’s going to be really interesting to see how everyone works on this. Like you said, I think we can approach it from different angles. We can think about improving what’s bothering us and helping to maximize the length of our healthy lives, as well as trying to address the root causes of aging and the degradation of our bodies, cells, and organs so that we can eventually extend that as well. There’s such exciting work going on in this area, and it’ll be interesting to see where it leads.