AI, Ageing And The Question Of Trust with Dr Filippo Dall'Armellina

Speaker A: 00:00:00

Foreign.

Speaker B: 00:00:05

Welcome to Beyond Longevity, the podcast that explores not just how we age, but how we can build a longer, healthier future for ourselves. My guest today is Dr. Filippo Dal Almelina, a researcher working across biotechnology, artificial intelligence, and drug discovery.

Biology is going through one of the biggest transformations in its history. For decades, scientists could spend years, sometimes entire careers trying to answer questions that today can be explored far more quickly.

AI is now helping researchers analyze enormous amount of data, predict the structure of proteins, identify possible drug targets, and speed up parts of scientific discoveries in ways that would have been difficult to imagine just a few years ago. Some scientists believe these advances could fundamentally change how we develop medicines and understand disease.

But despite all the excitement, AI is not replacing researchers. If anything, it is making human expertise, judgment, and scientific rigor more important than ever.

In this conversation, we explore how science itself is changing.

We discuss why so many promising discoveries never make it to market, why the relationship between academia and industry matters, how breakthroughs such as AlphaFold are transforming biological research, and where AI is already delivering results rather than simply generating headlines. We also asked some bigger questions. What can AI genuinely tell us about the human body? Where are its limits?

And are we sometimes confusing faster answers with deeper understanding? This is a fascinating conversation about the future of biology, the future of medicine, and the role humans will continue to play in both.

Hi, Dr. Filippo. Thank you so much for joining us today on Beyond Longevity.

For listeners who may be meeting you for the first time, tell us a little bit about yourself, your background, where you're from, what you've studied, why you've studied it, where you've studied it.

Speaker A: 00:02:26

Yeah, absolutely. So thank you so much for having me on the show. It's really a pleasure to be here.

I've been following the podcast, so I'm Italian and my background, my scientific background comes from research and I've done a bachelor's degree with honors on biological and medical sciences.

And then I got very intrigued and interested by what happens behind the scenes and research in general, which is why I pursued a PhD at the University of Liverpool. At the same time, I've also done some data training and machine learning courses in collaboration with UCL and a lab group over there.

But my main interest during my PhD was protein modeling and the field of structural biology, applied as much as possible to neuroscience and to the biochemistry that we are trying to uncover every single day. So I'm a researcher by background and I've also got. I've been involved in many projects.

I was involved with a project on studying the physiological effects of suborbital flights on astronauts with the European Space Agency. And I've also been involved with the vaccination program or COVID 19 with the NHS in the northwest of England.

So these are just a couple of projects that I've been involved with alongside. And along the way, one of my main focuses has been podcasting and I've been a guest on other podcasts and I've also got.

We've done episodes that have been a co host for Neuropod Cases, which is the clinical neuroscience podcast whose audience is mainly clinicians and people with professional medical background in that field. And so, yeah, so it's, it's very nice to obviously chat with a fellow podcaster.

As of right now, as I said, I have a background in research, but as of right now, I do the due diligence for startup companies in the biotech and pharma industry and I help with the scientific strategy that that is required for planning trials and designing new experiments and things like that. So, yeah, so that's a little bit about me and yeah, I'm very happy to be here.

Speaker B: 00:04:40

That's not a little bit about you, that's a whole lot about you. So I have to obviously sort of dissect all the different things you've said because there's so much to talk about.

But let's go back a little bit because you very elegantly glossed over something that I think is quite impressive and certainly worth mentioning. You did a BA and then you didn't do a Master's.

As far as I understand, you went straight into doing your PhD, which is rather unusual and I'm sure many people don't even know if that's possible. Just go back a little bit and tell us why you did this, how you did this. You know, just. Just indulge us a little bit.

Speaker A: 00:05:25

Yeah, absolutely.

I think it's definitely a one in a million case scenario in the sense that realistically I had two options at the time and two options I was interested in were graduate medical school and a PhD and research. But I was very intrigued by my honors project during my bachelor's degree, and that's a project that also became my first publication at the time.

So I was very enthusiastic, obviously about the being part of the field. And so I've obviously had to go through the same application processes and as, as everyone else.

But I think I was able to convince the board and the committee that I was probably the best candidate at the time just because I was very driven and I was very interested in the research that we were doing.

I had already read the EPHD proposal and came up with a couple of experiment designs and plans that we could have used as contingency plans for experiments that could have gone wrong. And as you know, in research, not everything goes as a straight line. So. Yeah, so I think that's, that's what happens there.

And I was very happy obviously, to make that happen. That's what I can say about that.

Speaker B: 00:06:35

Yeah, you're being very humble about it because I'm sure it wasn't easy. And as I said, I've never heard of anybody doing that, so that's definitely very, very impressive.

Do you think when you arrived in the PhD program, did you ever feel that you had skipped a step and that you sort of had to catch up quickly, or was it a pretty natural progression from your BA and then the research you did on your own to doing the PhD?

Speaker A: 00:07:04

I would say it was definitely a natural progression. I had spent the summer between my bachelor's and last year, the start of my PhD, during a summer research trip with the Wellcome Trust.

So I was doing research in the meanwhile anyway, and it ended up being the sort of like background research that I needed, then start my PhD. So I think it was the perfect setup.

I knew which lab I was going to work in already, and so I think that all the, the dots connected at the right time and in the right way for me, probably the most difficult thing during a PhD when you first start one is to actually know what the end goals will be and to, well, you know what the end goals will, will be, but you don't really know what's the sort of mini tasks and micro objectives you have to set up along the way.

I had been thinking about it for nearly a year and definitely for months, so it definitely wasn't a problem for me to actually put down in writing exactly what I had to do and to do it in a very systematic way. So, yeah, I think that's the thing that I would give as advice to anyone who decides to start a PhD.

Be very clear about what your objectives are and prepare for complete chaos, because some of those things may just not happen right away. It may take longer to optimize something, but you will get there eventually if you do it in a very systematic way.

Speaker B: 00:08:33

So do you think that chaos that you had during your PhD that helped you in a way, or was it more of a hindrance? Could you have done without it, or do you think that is part of what shaped you today?

Speaker A: 00:08:49

It was definitely part of what shaped me today. And it helped. And I think I couldn't expect anything else than just pure chaos, because science is pure chaos. And this is my perspective.

I mean, everyone thinks that scientists work in a very ordered air raid way, and we do, and we, that's how we plan experiments, that's how our analysis of our results actually works.

But then realistically, there's so many other things and there's so many other ways and approaches that a person can undertake in order to answer a single question.

And even if you've got really promising results in front of you, then there's probably another two or three limitations that you can discuss in that paper where you're trying to publish your results.

So if you want to obviously do all encompassing, if you want to write an all encompassing narrative regarding the results, you have to think about all the various limitations and that's just very chaotic.

So you probably, you can't do it on your own, which is why you have to start off with a PhD and you have to be supervised and work along your peers and postdocs.

But if you kind of embrace that chaos, and, and by chaos, I mean the fact that when you wake up in the morning and you go into a lab, things may not go the way you want them to go right away. So if you embrace that, then I think everything, all the other relevant steps will come along the way for sure.

Speaker B: 00:10:15

Did you always wanted to pursue a career in research?

I mean, I know now you're sort of doing both academia and commercial, you know, industry, but as such, has your goal always been to stay within research?

Speaker A: 00:10:31

Probably when I started my bachelor's degree, I didn't know that research was going to be my lifelong mission, but I've discovered it along the way. My lifelong mission has always been to obviously make a positive impact on people and in the world in general.

So I just didn't know how those and I didn't know how because I didn't have the appropriate knowledge and tools to obviously pinpoint exactly where my interests actually aligned and where I wanted them to be addressed. I discovered that research was a big part of my passion for science during my honors project.

In my bachelor's degree and during my honors project, I saw a lot of potential in scientists in general.

We have, and every single scientist has the potential to make a difference and to answer various questions that come to mind in a very specific field. And so that's very important.

I mean, aside from adding knowledge to what we know and adding topics and things that we can understand to Biology and medicine books, we can obviously help people because all of the information, even if it doesn't have a translational clinical application right away, will be useful for other people that will come along and need those results and those discussions that you've worked on in order to build new hypotheses later on. So I think research will always be a part of my life in one way or another.

As of right now, I'm involved more with the industry field, but I've got a researcher mindset.

So when I have to look into a startup or a clinical trial that a company presents to me, aside from obviously looking into the data room and looking at all the relevant information, I'm very keen on analyzing obviously how the, for instance, preclinical trials were actually designed, why they've done them in a certain way, and perhaps not with a different method, I know works better or worse. And so it's all about comparison in the end. I've still got that scientific and mindset in the back of my head for sure.

Speaker B: 00:12:39

Do you think researchers are too siloed? Do you think they're not enough in touch with the industry?

Speaker A: 00:12:47

Yeah. So I would say that there's definitely a gap.

So at the beginning of my PhD, I was seeing that the gap was huge and there was just no point of contact whatsoever between academia and industry.

But I think that I was part of the problem in the sense that I just wasn't well informed and I didn't really know what efforts are out there at the moment and what efforts were being made at the time in order to fill that gap somehow.

But then with time and experience, you get to know people and expand your network and you start to realize actually that there's a lot of efforts out there to bring that conversation between industry and academia, which is very relevant to, to everyone. Not every academic obviously is interested in having direct contact.

And I kind of understand that in a sense that not every single academic or researcher has the time or wants to spend the time to, obviously. And time is the most valuable thing in the world.

So obviously if they don't want to invest their own time into, I don't know, building connections or working with companies who build better antibodies because they need them in the lab and the very specific protein they're studying is not very well tagged or does not have good antibodies to work with, it's not like they're part of the problem, but it's their own decision to do that. But I'm starting to see sort of like a shift in more senior Researchers to welcome opportunities to collaborate between academia and industry.

And I think part of that, I mean, you hear a lot on the news the potential negative impact that artificial intelligence and deep learning can have to every people and every single person and sort of like what they do every day.

But realistically speaking, bioinformatics in general and what we're going to discuss in a little bit, probably, and computational modeling in general, I think has brought academia and industry a lot closer because now what we have is experts, so highly experienced experts like computational chemists and people like EMBA informaticians who have got the experience to be able to direct projects with startups and companies and tell them exactly where to focus in order to create better tools for them to do their own research and so improve their research output and to move the field forward as well.

So to be able to maybe skip a couple of steps along the way of just testing every single compound that presents in front of you and instead speeding up the process. And I think the value in that is immense.

Speaker B: 00:15:24

Absolutely.

Before we come to AI, I just want to talk a little bit more about the human side within the field of research, because a lot of researchers are very protective of their research and feel as soon as you're sort of talking to industry, you're selling out. You're being a traitor to sort of academia. How do you see that?

I mean, not you personally, because obviously we know you personally are all for it and are open to it. But about the field of academia, what's your thought on that?

Speaker A: 00:15:59

Yeah, what's my take? Well, I would say that I've definitely heard that before and I've heard things like, well, make sure not to move to the dark side of the spectrum.

And I think we have to be realistic here and actually put the cards on the table and, and look at things, at how they actually are. Academia works in a much slower way than industry does, and industry can actually help academia move it forward and move it forward in a faster way.

And there's been a lot of projects that were driven by the scientific community that then turned into initiatives on GitHub or initiatives in general that led to startup companies or ideas that were pitched to industry in general. That's the kind of communication and conversation that I think academics should have more often with industry.

So if they have an idea and they want to implement it, obviously companies in general, especially the most established ones and the largest ones, will have the power to make that happen, to set things up and to help you out. So how do you make those conversations happen? Well, I mean, first of all, you need the network.

So part of the problem of there being a gap between academia and industry is that it has to do with the framework we're working in.

Academics obviously talk to academics, and when they have meetings or seminars, they usually invite other academics and people from their own universities or other institutes. So these are things that can be potentially easily fixed.

I mean, if industry shows an interest, obviously to participate at these meetings, there obviously has to be someone within the company says, well, I will spend time doing this because it potentially has value. Then you start creating a platform where academics can reach out to members of companies or representatives in a much easier and more convenient way.

I definitely do think that we've moved forward a lot compared to what things were like maybe 10 to 15 years ago, where academia was mostly doing its own thing and industry was just running in a parallel line just along the way. And so it was a two way street, but there was no point of contact whatsoever.

Whereas now I see also a lot of academics posting on LinkedIn and posting in general and talking a lot more about the people they've met in big pharmas or even small startups. And that changes everything.

And now what we see is that there's always more academics who are like full professors or senior academics who are getting more and more involved in the scientific advisory boards of startups and companies.

And obviously helps too, because their full time job remains in academia, but they kind of become the bridge that fills that gap that we were talking about. And so it makes it easier for them than bringing colleagues and create conversations that can be meaningful along the way.

Speaker B: 00:19:04

Yeah, absolutely.

I've had a few researchers on the podcast and I think a lot of them are really surprised that what they have been researching on is actually directly applicable to the world out there.

And I think a lot of researchers are surprised at the direct impact they have because research used to take such a long time and often the research would only come to fruition when the professor working at it or the researcher was sort of old or has even died. So that has definitely sped up. And of course AI has done its own to that. So we'll come to that in a minute.

But why do you think so many promising scientific discoveries fail to become meaningful therapies or products?

Speaker A: 00:19:58

Part of the problem is that is definitely an admin and an organizational side of the things. So obviously organizing the experiments and planning the trials as they are supposed to be, to be planned, it's not easy.

So the role of people in the startups, however, well, they may do it. Obviously the therapies themselves will not necessarily work a hundred percent of the time.

Actually we know that most therapies that get to to phase one or to phase two will fail. So that's definitely one thing.

And part of why some initiatives and some projects may have failed has to do with the speed that two different world wards actually work in.

So academics will have definitely less time to to prioritize their involvement in certain projects with startups, whereas obviously the startup is betting a hundred percent of their resources or whatever percentage that is into a specific project alongside them. So that's definitely one problem.

Obviously if for instance, I know cases where advisors like academics, if they had spent more time actually looking into all the potential things that could have been screened perhaps during IND enabling studies, they could have maybe looked at the extra information from toxicity studies that would have tell them, well, maybe this actually shouldn't be brought forward to a later phase along the clinical stages in pharma because it doesn't really show any potential. We have to go back to lead optimization or that's definitely part of the problem for sure. How do we make that better?

Well, that's a really good question and I think everyone has different opinions on this.

And realistically speaking, I think people who come from a research background, if they're part of the company itself, then I mean, when I see researchers being part of the company themselves, Maybe they've given 100% of their time doing that and they still carry out the communication and conversations with academics, that's a good sign to me because it shows that there's someone that's part of the startup that actually believes in what they're doing and maybe was working on the field as well before that as a researcher. And that I think adds extra value because they will know the field a lot better and they will know what to look out for.

Especially these days when I see so many startups coming up with various projects and ideas.

Some of them, if these ideas had come around the time when I was starting my PhD, maybe the projects themselves would not have been carried forward in any sort of way. I mean, just for instance, we now know that mitochondrial health is very important both in the longevity field and in biotechnic in general.

And there's so many startups now that are targeting various compounds and various proteins and genes that have to do with that field.

But I think a while back everyone would have just said, well that seems like a niche, like a very small subfield that will probably not have a large impact but now we're starting to see really good results in phase 2 trials of clinical trials where the targets have to do with mitochondrial health or transport of proteins. And so that's very important as well. And that's probably why the basic research we are doing today does not look as it can.

It has a lot of translational potential. There might be translational potential in clinical applications in 5, 10 or 15 years.

Speaker B: 00:23:37

Yes, definitely, because it's all sort of speeding up and we don't know where one thing will lead and what will come of it. And a big part, as you've mentioned, you know, already a couple of times, is AI. So we have a lot of listeners that are not researchers.

You know, they don't know much about biology as such. They investors. A lot of people hear about AI every day, but far fewer understand how it is being used in biology.

What is AI actually changing in the research field and within biology at large? And what can AI do today that generally wasn't possible 10 years ago?

Speaker A: 00:24:21

I think we have to consider what the hype is these days, because there's immense hype regarding the applications of AI in medicine and in biology. But we have to be careful about exactly what AI is doing and what it can do, and have to be realistic as well about what the prospects will be.

So in order to put a little bit of clarity in this topic, I think I should say that. So I come from the structural biology and computational modeling field mostly, so I can see what the applications of AI are within these fields.

So just to be clear, for anyone who's, who's not an expert in structural biology is the field of science that studies the three dimensional shapes of biological molecules, particularly proteins, and how their structures determine their functions inside living organisms.

So because proteins carry out most cellular processes, understanding their structure is essential, obviously for explaining health, disease, and how drugs work as well. And computational modeling is part of this.

So computational modeling aims to predict and simulate these molecular structures and interactions as well using computer algorithms.

This is a field that has existed for decades, and only now we're hearing about what the potential applications are, because only now this bubble has kind of exploded.

And in fact, in recent years, artificial intelligence has just revolutionized this area by enabling highly accurate predictions of protein structures and biomolecular interactions directly from sequence data. So the sequence is the, is what we use in order to identify what proteins are.

So proteins are made of amino acids, and what we will see is a sequence of amino acids one after another. And these advances in AI are helping researchers better understand Biology and accelerate drug discovery.

And they can also help investigate complex processes such as aging and age related diseases.

But we have to be clear here, because the AI revolution in structural biology was only possible because of decades of experimental science that is now summarized in places like the Protein Data bank. Without the PDB, we wouldn't be here today. So advances such as AlphaFold that people will probably have learned of because of the Nobel Prize.

So these advances were built on enormous data sets generated through X ray crystallography, which is also something I've worked on before, and NMR spectroscopy, and also there's other structural biology techniques. These are probably the two that people will have heard of in the past, a lot more.

And so because of this, we have to be careful and highlight the critical importance of high quality experimental data. So if the experimental data is not that great, then obviously our model will not be as good either.

So the experimental validation to this day remains essential. In the age of AI, there's no AI without experimental validation, and we have to keep moving that forward.

AI will not give us all the answers right now. If we keep moving the experimental validation along, then AI will probably have a few more answers to our questions in a couple of years or more.

So machine learning models can accelerate the discovery, but they still depend on new experimentally determined structures in order to improve their accuracy. And so without these data sets, the protein structure prediction just would not have reached its current level of accuracy.

And As I mentioned, AlphaFold is one example.

hy it was recognized with the: 2024

And we have to understand that what the Alpha Food 3 is doing is that it's basing its information from multiple sources of information. So we've mentioned the protein sequences, it's one thing, but there's also evolutionary conservation information.

So this has to do with what those protein sequences are among species and over time as well.

And the co evolutionary relationships as well between amino acids is another very important thing that is considered by the model known structural motifs that we've discovered over the last 100 or so years, and also the physical chemical properties of proteins and lipids, so fatty acids and, and any other molecule that is within the cell. So one fascinating concept is that proteins evolve together and they often interact together.

So another important thing is to be able to study them not just in a isolated scenario, but in a sort of scenario where there's other proteins around and you can form macromolecular complexes.

And that's why by analyzing the evolutionary patterns, so it's, it's, it's, these are just patterns that the AI is trained on recognizing and then predicting. And the AI models can infer basically which amino acids are likely to be close in a three dimensional space.

So it's important to, to know that AI is not deterministic, it's probabilistic, but just like the rest of science is as well. I mean, so there's been loads of cases where we saw great results of drugs working in phase three trials.

And then once they came to a later stage and they were tested on different patients, they actually had adverse effects. And so that goes to show that the science behind it was done very well, but it wasn't deterministic.

The AI will not give you a true or false answer or a right or wrong, but that's why scientists are still needed and they will always be needed in the lab because they, they are the ones that can actually look at the science and look at the data and actually analyze it and tell you and tell you how to interpret it. So a common misconception I think is that AI predicts a structure and it just gives a definitive answer and that's not the case.

And this is where confidence scores become critical and this is part of the work that I've done back in the lab as well. So confidence scores basically tell you how confident we can be about the structure that we are looking at.

In practical terms, confidence scores can help researchers prioritize experiments and identify flexible regions and compare competing structural hypotheses, or like evaluate protein ligand complexes and then benchmark everything. So this is I think, an area, an area where, where future research has to focus a lot more.

So there's no point in having a lot of predictions and a lot of models if we don't know how to interpret them. And scientists are struggling to interpret the results as we have them today.

So there are peer driven and like scientific community driven initiatives to create new confidence scores that are better and that not only analyze the positions of amino acids and, and the interactions of proteins, but that look into other mathematical and statistical properties that are important during the model generation process. But I think we still are early stage and so there will be more to see, I think, in the foreseeable future.

Speaker B: 00:32:15

So before I ask a little bit more about how much we should trust AI and all that. I just really want to pick up on something that is been quite a breakthrough in the field of biology over the last few years.

I think it's the biggest breakthrough and it hasn't come from a Lab. It is AI and it is what you've mentioned already, AlphaFold, because you've already said that.

But again, I just want to make it clear to people who are not too sure about, you know, what it is and exactly what it does. AlphaFold suddenly made it possible to predict the structure of proteins at a scale that just hasn't been possible before. Is that correct?

And why is this such a big story and breakthrough?

Speaker A: 00:33:05

Yes, that's correct. And this is a huge breakthrough. I think it might even deserve a couple of more Nobel Prizes. No, I'm just kidding.

But what actually has changed is that the before, what we had to do in order to understand the structure of a protein, and understanding a structure of a protein is very important for understanding what the biology is, because, for instance, if we discover the structure contains transmembrane domains, then it means that those domains, they sit within membranes of organelles inside the cell. And that means they're probably not going around in the cytoplasm.

And so I'm probably being too specific here, but that's just one example of why this is important. And then understanding what drives the interactions between proteins, that's also another very big thing.

So is it the electrostatic potentials or salt bridges or what chemical modifications can drive that interaction, for instance? So these are all very important questions, and one that probably Big Pharma is most interested in is, does, does this protein have pockets?

Basically, is this protein druggable in some way or another?

And so pockets are places where obviously smaller molecules can sit and they can potentially change the conformation, the shape of the protein itself, which will, obviously anyone with a background in biology will know that structure and function have a direct relationship. So if you change the structure, then the function as well will likely be affected. And so what has AlphaFold changed?

Well, before, what we had to do was try to get crystals to basically then shoot X rays at the crystal and get diffraction patterns in order to be able to predict with mathematical models what the structure of the protein actually was.

The bottleneck here is that it's really hard to get crystals of proteins because what you have to do is you have to purify your protein and you have to make it form basically ordered arrays.

And so you have this protein in large amounts and all of the, each Single unit, each single molecule has to sit in the same exact position, in same exact orientation, so that then when you shoot the X rays and you look at the diffraction pattern, you can infer exactly which points correspond to what amino acids later on. And this, this happens also these days, mostly via molecular docking. So docking computationally predicts how a ligand binds to protein.

And so the goal here is basically to estimate the binding orientation, the binding affinity, through various interactions like the salt bridges I mentioned, or hydrogen bonds or hydrophobic contacts. And so there's already software, and there's already been software for decades that has been able to obviously do this.

But in order to get the crystal and to purify the protein, you have to do a lot of attempts and it can become a very long process. What AlphaFold can do now is just replace.

Well, it doesn't replace, obviously the answer as of, as I've said, it's not deterministic, but it replaces the entire pipeline by just using the information it has ready from all these crystal structures and all these experimentally validated information. And it predicts exactly what the confirmation of a sequence you give it, of a protein you give it actually is.

It doesn't just use the sequence information as we've talked about, and you can do that as well for complexes. So basically models where there's more than just one protein and where there's maybe fatty acids as well.

And that's, as you can probably imagine, that has so many applications in pharma and in drug discovery in general.

Speaker B: 00:36:58

You've mentioned predictions a few times. What is the difference between prediction and understanding?

Speaker A: 00:37:05

Well, prediction is based on the computational model that we are looking at. So, so it uses the information that you train it on.

So AlphaFold has been trained on so many models, and what it uses is it infers basically what the patterns from the training data are compared to the patterns that you are looking at and that you're testing.

So the protein you have in front of you, and so when you think about it, it doesn't actually know what the true structure is, but it gives you a value, a confidence score from zero to one.

And that's a score that it predicts and it simulates itself and it calculates based on how it thinks that it has done well or not at modeling the protein.

So if there, there's not many patterns out there and there's, there's nothing that it can base the, the prediction on, obviously the, the score will be lower.

So the prediction is based on scores and metrics of how confident we can be that the structure we are seeing does not have as many errors as it could have. Okay.

And so whereas our understanding is when we have a final product, when we have a model that has been validated experimentally, for instance, so a crystal structure, we understand that that protein has that very specific structure, but only in that very specific time frame. So that that structure is, is like taking a picture of a protein. And when you think about it, we talked about chaos before.

Inside the cells, there's pure chaos. So there's, it's, it's an environment that's filled with organelles and proteins that are just dynamic and moving continuously everywhere.

And so realistically, there are more rigid and more flexible proteins. So it's possible that that crystal structure, perhaps of a rigid protein, actually shows you a very native like model.

But the understanding to us is not just using the prediction and it's not just using the crystal structure, it's putting it in context.

So it's using the information that we have from the structure, looking at the pockets, looking at what we know can interact with the protein using other experiments.

So not using crystallography, for instance, by using Western blots or immunoprecipitations or any other kind of wet lab assay or microscopy as well, in order to be able to see whether proteins are close to each other, whether they interact and in what cases they interact. And to actually put that into context, into context with the structural biology gives us an understanding of what's happening.

And that's something that AI can't do today, but not that scientists can do, and we've always been able to do it.

Speaker B: 00:39:54

Does AI actually understand biology or does it simply recognize patterns?

Speaker A: 00:40:00

That's a really good question. And I think that arguably, perhaps a different guest or a different structural biologist might give you a different answer.

But in my opinion, it understands the patterns, so it looks for the patterns and he understands what its job is and then tries to experiment into the new project that you've given it. And it does its best job.

It understands only part of the, or really only part of the picture, I think, because the information that it's been trained on and the way that it's been trained, as we mentioned, is by learning from multiple sources of information, like protein sequences and evolutionary conservation. And we've talked about the physical chemical properties, but there's probably so much more that could be built into all of this.

For instance, AlphaFold doesn't use information from binding affinities. There's other AI tools that instead Put a lot more weight on predicting models using the binding affinities that are known experimentally.

And so you have to give it an input of sources of information that it can use to. To predict the structure. It will not understand the biology completely because probably we don't understand the biology completely.

We can infer things from the biology that we have in front of us, like the experiments that we've done. But yeah, so that's my take on this.

Speaker B: 00:41:27

You know, this talk about AI is just so fascinating. And I think we could have not just one whole episode, but 10 whole episodes just on AI and the development of AI within biology and all that.

I want to move along a little bit and ask your thoughts on the field of longevity overall. Where do you think is the biggest disconnect currently between science and the commercial longevity market?

Speaker A: 00:42:00

Wow, that's a great question. So the biology of aging is extraordinarily complex. I mean, we know about the hallmarks of aging.

I think the listeners will probably have heard of them before. Well, definitely in your episodes. But identifying actionable therapeutic targets remains difficult.

So I think the connection between AI and longevity is pretty straightforward here in the sense that AI can allow us to integrate the large biological data sets. I'm talking about data sets that are just so big that probably not in my entire lifetime I could navigate them by myself.

It can also allow us to discover novel targets. So I think this is where it can sit.

It sits in this space of modeling biological pathways, predicting protein interactions, and designing molecules more efficiently. And why is this important? Because this helps us prioritize experiments.

Obviously, the longevity field has received a lot more interest in the recent years.

And so the same way that AI can help big pharmas do what they were doing in the past already in biotech, it can help longevity startups as well to address the same points in a similar way. So, I mean, a particularly interesting example that comes to mind is radio Life Sciences.

So the Rubedo Life Sciences is a company that uses AI enabled discovery, an AI enabled discovery platform called Alembic, and uses it to identify therapeutic opportunities that are related to cellular senescence, which is, as we know, a hallmark of aging. We've talked so far about the fact that computational modeling can help big pharmas, and bioinformatics are very important.

There's also been advancements in artificial intelligence in other fields of bioinformatics, and not just structural biology. There's been advancements in spatial omics.

And so all of the information that, for instance, we can have from DNA sequencing and RNA sequencing of actual cells. And that's what the alembic platform of Rubedo does.

So it is designed basically to identify pathological cells so what, people may know them as zombie cells that drive aging and chronic diseases, and to basically engineer drugs that can selectively target them. So the AI integrates massive data sets here. As we've learned, we need these massive data sets.

And it uses massive data sets from single cell RNA sequencing, spatial omics, and patient clinical samples as well. And what it can do is that it can distill and condense the complexity by mapping out the unique molecular signatures of senescent cell populations.

Then machine learning algorithms can sift and course through this data to identify exact subpopulation of pathological cells that drive the disease progression, along with their unique therapeutic vulnerabilities. And once the novel targets are identified, the pipeline then uses medicinal chemistry to design adaptive xenotherapeutics.

And these are highly selective small molecules, including pro drugs. And they are engineered basically to either destroy deeply damaged cells or restore homeostasis in salvageable cells.

And so by utilizing this pipeline, we beto advanced their clinical lead candidate, which is a GPX4 modulator, into human clinical trials for dermatological and inflammatory conditions. And so there is data on actinic keratosis, which is a precancerous skin condition associated with chronic sun damage.

And obviously, as we age, we are more exposed to sun damage. It also depends on which part of the world we live in.

But this data basically comes from phase 1b and phase 2 studies, and it basically shows, and they've published them recently, and it basically shows encouraging reduction in lesion counts together with favorable safety profiles as well. So this is just one useful example of how AI assisted discovery can help identify novel targets and then accelerate movement toward clinical testing.

And Rubedo also illustrates an important point about the current longevity biotech landscape. Because of this, they are targeting cellular senescence, which is a homework of aging.

I'm sure that there will be other attempts to do things like this and to perhaps address multiple other hallmarks of aging at the same time, like inflammatory aging and other hallmarks as well. And this is where the science is moving forward. This is how AI can probably help the longevity space and help fill that gap that we can see now.

Speaker B: 00:47:04

So what do you think needs to happen over the next five years for you to say that AI has genuinely transformed longevity medicine?

Speaker A: 00:47:13

I think that the potential is huge, and I probably will not be able to just summarize every single thing that can happen, but there are Interesting developments. And I think if we see these, these developments taking place, then I think we're getting closer to answers.

I mean, I saw recently a post on LinkedIn that was saying that Forbes recently showcased Kai Discovery, which is a, an AI drug discovery startup. And what they do is bioinformatics and modeling, similarly to what aptfold does.

But Pfizer and Lilly will get early access to kaifree, which is their latest model, to basically help accelerate drug discovery. This is a notable partnership.

I mean, we're talking about the largest and most established drug discovery organizations in the world becoming more interested in actually spending, I think in this case, over a billion dollars on, on AI driven drug discovery. And so what do we need? We basically need more collaborations like this to also be able to see to what extent AI can actually help us.

So if collaborations like this unravel new biological pathways and help us discover new potential drug targets and potentially also create new pro drugs that can be tested clinically and that can also reach the clinical stage, that's the very important part in this, then I think we're moving forward in the right way. What we need from academia in the meanwhile, and this is probably where industry can help as well, is like I said, improve our datasets.

So we have huge data sets of information of, for instance, structures of proteins. We can actually make those a lot bigger.

Instead of having academics applying for small grants to be able to study just one single crystal structure, we can probably do this more systematically. And there's a lot of platforms and infrastructures like diamond out there that can help do this and they are doing this.

So it's important to know that when the funding has to go in both directions and not just follow a single track. I would say now, not to end.

Speaker B: 00:49:29

On a negative note, but we've looked at the future now. What do you think when people look back on this period now, in 20 years time, what do you think we will have misunderstood most about AI?

Speaker A: 00:49:46

I always like hearing other people's opinions on this as well, because everyone will have something different to say. And I think that more or less we're all right in saying that we have to be careful about how we interpret what AI tells us in science in general.

I would say that looking back, I can see how perhaps a lot of funding would have been wasted or gone to waste in creating new AI platforms or creating new AI initiatives when the ones that we have today that are perhaps, that are perhaps an initiative that came from academics and from labs are just okay. And perhaps using that funding to Improve the datasets and to improve the model itself.

Improve something that already exists means using our resources a lot better.

So I think that that would probably be one thing because I can see that there's a lot of hype and people wanting to be part of this hype are trying to do the wrong thing. But we don't really need to see a competitive landscape.

We just really need to see one thing that works for everyone or, or a couple of things that can be benchmarked against each other so that we can use them to infer exactly which drugs are the best drugs that we can move forward from lead optimization to then IND enabling studies and so on. So yeah, I think that would probably be one thing I could expect looking into the future.

And also one thing I'd probably expect is that, and I can see this happening with AI agents a lot that we, we're starting to rely a lot on AI and up until the other day we didn't really need it in the sense that we were able to carry out our job, maybe in a slower way, but. And perhaps missing out on some points that AI helps us uncover. But relying on something new so much so fast can sometimes be. Be tricky.

I think we have to be realistic about what we have in front of us.

So when we publish new articles or when we do new research, it's important to have other science, other methods that actually complements what our narrative actually is.

Speaker B: 00:52:07

I think, I think you're absolutely correct what you've just said and I really, you know, I agree with it 100%. It's very important not to rely on AI exclusively and 100%, and yet we need AI to advance what we have.

As with everything in life, I think, you know, it's a fine balance, but an important balance that I think we need to strike in order to move forward in a positive and correct way.

Speaker A: 00:52:43

Yeah, absolutely. Like you've said, perhaps setting some rules or agreeing on how we should move forward from here on would be a good idea.

I think this has to do with all of us. So every single person in the field should probably ask themselves the same question. What can I do to make this happen?

Speaker B: 00:53:05

Very wise words. So now you know, we're at the end. It was so interesting and so informative. I ask all my guests some rapid fire questions. So no exceptions here.

What's the single best piece of advice you would give your younger self?

Speaker A: 00:53:22

So I would tell myself to be clear about my research objectives. There are so many things that interest me and that have interested me over the years. So just be clear about what your projects should be.

Time is obviously the most valuable thing in the world, so make sure to use that properly.

Speaker B: 00:53:43

Name one habit everyone should adopt for a longer, healthier life.

Speaker A: 00:53:47

One habit everyone should adopt is definitely, in my opinion, to have a hobby or a passion other than what you do every single day, that motivates your mission. So it can be meditating, it can be exercising, anything, but just have that with you.

Speaker B: 00:54:03

If you weren't in longevity science, what career would you have chosen?

Speaker A: 00:54:08

Aerospace medicine has always interested me, but that's also because I've always been interested about space travel and astronomy in general. So perhaps I would have been an astronaut.

Speaker B: 00:54:19

You know, we've had a doctor on who's been. Who was a doctor for Ezra.

Speaker A: 00:54:23

Yeah, so I've. Yeah, I've seen the. I've listened the episode.

Speaker B: 00:54:26

Yeah, it's very cool. You're saying. Anyway, so what? Microdose habits, five minute routine or small daily action yields outsized longevity benefits.

Speaker A: 00:54:38

One of the best habits would be to eat healthy as much as possible. So I'm not saying that you have to follow a very strict diet, but if you can follow at least that 80, 20 rule.

So the 80% following, for instance, the Mediterranean diet, which is shown to improve healthspan and health in general, then that's great.

Speaker B: 00:55:00

What's the craziest longevity myth you've encountered and is there any truth to it?

Speaker A: 00:55:06

Well, one of the longevity myths that you hear about sometimes on the news is that avoiding plastic as much as possible improves your health span and lifespan. We need a lot more studies, I think, in order to be able to know that field a lot better, in order to make claims like that.

I mean, surely ingesting plastic isn't good for you, but. Yeah, you see where my. You see where my point is?

Speaker B: 00:55:33

Yes, often things are a bit exaggerated. There's a good basis to it, but people take it out of proportion.

Speaker A: 00:55:40

Yeah, absolutely. Touching plastic will not enhance your vas, your reactivox, which pieces and. Yeah, so just be careful about what's out there.

Speaker B: 00:55:50

Thank goodness for that. Effy. Thank you so much, Dr. Philippa. It was riveting, this conversation. I had a great time.

We dove into so much detail and as I said before, we could have filled five episodes just on AI alone, let alone all the other wonderful things you're involved in and you're doing. So thank you much, so, so much for taking the time and coming on beyond longevity.

Speaker A: 00:56:16

No, thank you. It's been real fun and I really enjoyed it.

Speaker B: 00:56:21

Artificial intelligence may become one of the most powerful tools medicine has ever developed. But as Dr. Filippo explains, a tool is only as useful as the people using it.

AI can analyze vast amounts of data, identify patterns humans might miss, and accelerate parts of the scientific process. What it cannot do is replace good science, careful thinking, or the judgment that comes from genuinely understanding biology.

The most exciting future may not be one where AI replaces scientists. It may be one where scientists equipped with with AI can answer questions that would previously have taken decades to solve.

If you enjoyed this conversation, please follow rate and review. Thank you.

Speaker A: 00:57:28

Sa.