#420 Dan Jamieson from Biorelate: How AI is Discovering New Drugs Faster Than Ever

0:00Welcome to the Health Tech Podcast. Here we talk about everything healthcare and technology. and I'm your host James Summery. Hey really, this week I'm joined by Dan, the CEO of Biorelate, described as the most advanced data curation in biopharma with a product called Galactic AI, which all sounds very grand and wonderful. So Dan, welcome to the Health Tech Podcast. How are you doing? Doing very well, thanks for having me. I've been looking forward to this one, James. Yeah, it should be fun, man. It should be fun. So I think it's probably worth explaining, first of all, to listeners, the value of, well, I'm going to tell this from my point of view, the value of going to events, going to things like dinners and just speaking to the people sat next to you.

0:50Because we were first sat next to each other. What was it like four years ago? Three years ago? It was a long time ago, wasn't it? We went to, we sat next to each other at this dinner, added each other on LinkedIn, and then have just been following each other's careers, seemingly. And then at some point in the last few weeks or months, we started chatting again and started doing some work together. And now we're here, we are on a podcast. So the value of serendipity, but the value of the value of what was it someone said to me the other day? The value of community or the price you pay for community and the price you pay for serendipity is sometimes making yourself uncomfortable.

1:26So getting up off the sofa, getting dressed and going to a dinner is probably a good thing sometimes. And that's the price that you pay for some value in the world. But yeah, good to see you again, mate. How are you? I'm very well, thank you. And it's good to see you too. So Dan, if you saw for our listeners, if you to tell us your story, because I think to get to a point where you're running a company like yours, there has to be a I mean definitely a level of education level of experience and then a level of kind of I don't know like ambition and dreaming up this idea out of nowhere and then and then knowing how to execute on it like you don't just wake up one day knowing how to run a company like this and so how how does one amass the skill the knowledge and yeah I guess the impart the belief that you can run a company like this?

2:19Like, where does your story start? Honestly, I think it starts with having a tiny little bit of each of those things, education, the belief, and to some degree, some knowledge of running a company from, you know, what you've learned from other people. But, you know, I'm a first-time founder, so BioRelate was my first company 11 years ago. And when I started it, honestly, I didn't really know what I was doing. I was really doing a lot. And when you start your first business, there might be some people out there that say, yeah, I always knew I'd be blessed for and I had every ounce of belief that I would get to where I am now, et cetera.

3:05That's usually, I think, people talking with hindsight. That's a polite way of putting that, by the way. I mean, if you look back, So I started my company in a PhD, right? And the environment that I was in, nearly everybody there was working towards doing a postdoc, or at best, thinking about a career outside of life sciences and into a well-paying job. some people were also obviously going towards industry but most people just weren't thinking about entrepreneurship which actually pretty unusual in the states in the uk we're far less entrepreneur and our phds whereas in the us i think people there is a bit of an expectation that you can go out and spin out a company so um yeah it was it was it was an environment where entrepreneurship was not the norm um and this was um yeah it just made that you were fighting against all of the expectations um to just become like a postdoc or or just do something very routine and normal so starting a company was definitely almost um a contrarian at at the time it was it was seen as something that you um you were less likely to be successful in compared to all of those other options and i um i kind of just threw myself into it and where did it really start i i think it started with wanting to um wanted to build something myself and wanting to um yeah be in control of your own destiny that people often say that don't they so i think that was where where the whole drive began from but then it was slowly just building up confidence and building up belief by doing one small activity after another that was like a success point some kind of measure of progress so for me in those early years I was obviously doing research in a PhD and that ended up being the foundations for my company so where I actually really started was I went to Pfizer they were looking at developing novel drugs for chronic pain diseases and in my PhD I was researching or I was using what we called it text mining or natural language processing and today everybody just calls it AI or generative AI.

5:38When was this then? What year was this? This was 2013, something like that. Oh wow. Yeah, so really was the precursor to what most people are using in NLP and AI today. yeah okay um so i was using some much more primitive technology to try to mine literature for for data and knowledge that we could help uh drug discovery companies like fisa in this case um have better understanding of how drugs work and so what fives were trying to do was trying to find novel targets for chronic pain diseases that they could then go off and do a load of research and progress onwards through that clinical trials journey and so I had this approach which mined the literature for cause and effect so relationships between proteins chemicals drugs drugs um biological entities that once you you gathered all of this data and stitched it together into networks you could then begin to understand the mechanisms behind how these diseases and drugs might work so we built and we called this the world's first pain interactome which really doesn't sound very sexy pain interactome exactly it was um yeah um marketing wasn't my skill back then so we called it the world's first pain interactome and actually i think the publication was was called something like mining the biomedical world of pain so we we really were just just going for it at that point in terms of bad marketing um and yes we once we assembled this thing, we then use that to essentially propose a bunch of novel drug targets that Pfizer had yet to experiment on.

7:37And they already had a load of previous screening data that they could then immediately compare against the targets proposed. And many of the targets proposed looked... There was a trend towards the ones that they found as being hits within their screening data already. So they've done some pain phenotypic screening data. So anyway, they went off and did a load of new experiments on the ones that which they hadn't yet to have worked on. And they ended up doubling their hit rate. So it went up to something like 40 % from those proposed targets. Yeah, which is a really cool result because the way they were doing it previously was having world leading pharmacologists at Pfizer come up with those hypotheses themselves.

8:29So this was a really nice proof point. And because this was in an industry setting, you saw the value of it as well. You saw that Pfizer then wanted to continue on with those experiments to see if they could progress those through to the next vehicle trial. So this gave me my first taste of what success could look like if you were driving it with a product. in this case network that we've built for pain and dan just to jump just to jump in a sec just just before you move on so you are mining the you built a tech product that mined the literature for cause and effect so you can start to understand potential drug targets and so if I've understood this correctly Pfizer then had some data on the things that were proposed some of the things that were proposed but they didn't have data on others and then when they started exploring those others they've then realized oh hold on a minute there's some things that could really work here now that's really in it's really interesting isn't it because I think you hear this phrase a lot the things like you know the future's already here but it's just not well distributed or you know phrases to that effect and we we see it in health tech as well as biotech that there are areas where technology that we already have or in this case knowledge that we already have can do so much and i think it's interesting that the technological frontier here is not actually creating new information it's surfacing new information to some people but it's actually just backing up this amazing industry that we already have in scientific research and academia and adding utility to it which i just think is interesting it's interesting that the whole point of that scientific research and literature is to surface it to everybody but it doesn't work without this thing that you've built perfectly if you know what i mean when i when i often explain what biorelate does um so we'll obviously get to this when when i'll tell you the full story the cut end and how we're adding value today but it's very similar to that um to that pfizer example in that you have all of this knowledge that sits out there in the biomedical domain in publications in clinical trials it's unstructured um yeah but there's this um this mistake that people making thinking that it's just the case that people haven't read it yet and therefore you're helping surfacing existing information to people so they're fully aware of what's out there which is true that is a problem but that's not the that's not the the full value you get from this approach What people miss is that once you stitch together this information, you get multi-hop-like inferences that you can make, which improve your understanding from what was posted directly within any one piece.

11:46So to give you a really discreet example, if you were to take a paper that published one relationship between, say, a protein phosphorylating another protein, protein A phosphorylates protein B. um and then you know 10 years ago in an obscure journal and in you know someone in non-beady bread or something like that somebody said that this phosphorylated protein um once it's phosphorylated causes hypertension or something like that right um so you've got two different pieces of research two different causal relationships two different effects um but you stitch those together that gives you an insight that protein a may cause hypertension once it's phosphoridated protein b um but there's no there may be no individual piece of research that's talking about protein a influencing hypertension so you actually get new information and new knowledge from stitching together this stuff and so i'll give you a really really discrete example but if you think about how this plays out in very broad networks and once you apply you know some decent statistical tests to how to analyze this this kind of information it becomes very very interesting very very quickly and so what you got from from this example of building a network in pain for Pfizer it doubled the hit rate because it was giving unbiased hypotheses that these people um the experts in pain with all of their existing knowledge of pain um weren't able to garner because they hadn't seen it before right they probably did have most of the the key knowledge surface within their brains already um most of these hypotheses were actually the kind of hypotheses that didn't make sense the the people who were so you get you get these really really interesting examples once you stitch this information together.

13:46But I think the point that you were making as well is that there's a lot you can do with existing technology at them before we had generative AI and everything else to actually add tremendous value if you know how to apply it. And I think that was one of the most important insights I got from that particular Pfizer project because otherwise I was in a PhD building technology to publish papers which i mean about the outcome that we use most people seek within their their phds they they do their research they they get a thesis or they if if you've done well you get you know a few publications or something like that whereas if you're doing it for industry you're doing for pfizer and that actually contributes towards um the potential to develop pain drugs um that's a lot more valuable and you're actually really are doing something that that has an impact.

14:41So, yeah, I think it's a case of finding the right application for some of these really interesting ideas. And a lot of people are trying very, very hard with the generative AI technology that we see on the markets today to find these really pointed examples of where it can add value because a lot of money has been spent on these technologies now, either in how they've been adopted within companies or all the companies themselves trying to show that they're going to make a profit from all the billions of investment that they're taking in today. Yeah. So what happened next then? Because this is it's a heck of a light bulb moment, I suppose, that you've built this piece of technology that all of a sudden and bear in mind, you know, the drug discovery pathways and at large, there are so many companies And it's such a massive industry of people trying to make 0.1 % of difference to this and to reduce costs by 0.05.

15:43Like it's so small margins in terms of the marginal gains that people are trying to make that doubling the hit rate of proposed targets seems like a really significant jump. So was there a, did you realize the value of it at that point? Were you told the value of it at that point? Were you given the energy to pursue this by signals coming to you by this point? Because I've noticed in the world that when you do create something of value, there's a lot of people that start circling you at that point. So there's a lot of interesting IP locked into this PhD project at this point. You know, it's a really, really interesting question.

16:27So I think it certainly was valuable to Pfizer because they went ahead and progressed some programs based off of that analysis. It happens over a long period of time, though. So I made that that story sounded like it all happened in a couple of months. But each of those analyses, each step afterwards, it takes a couple of years for this stuff to properly progress. Right. So, yeah, it was kind of hard to see the value in a quote unquote light bulb moment. But you kind of got the sense that this was was important. And also because you're a PhD student and you don't really have much business sense, you don't really understand the value of what you've created either.

17:11So somebody, our advisor might have understood the value of this, somebody that was working on budgets and providing some kind of impact statement for the research they were doing. But as a PhD student, your currency of success, again, is the publications that you get from doing this kind of stuff, which in this case we did. we published all of the findings that we got from it. And it became actually a really useful piece of collateral once I had started Biorelate because I had published proof points to show that this approach actually does work. But it didn't materialise to me immediately that this was immensely valuable um this was something i actually learned as i was building biorelate that the true value of doing these kind of things um it's only when you go out there and you you have to sell something to somebody at a price that you realize the value of things and back then there was nothing that was being sold it was literally just me as a phd student doing some research sure um so what was the next step then after the Pfizer project into kind of turning this into a business um so I think then it was just um a very haphazard journey of um of trying to get started um so I suppose the things that made a difference were I entered a few startup competitions and I won a couple of them which gave me a bit of confidence um one actually gave me something like 10 000 pounds in in winnings which as a phd student that's quite a lot of money that's huge yeah that's huge so that was really i think enough to give me a sort of kick to go off and start the business not get a job and have a go um but i should um i should also say just how poor it was back then it was uh most people have like some kind of savings or some um you know some something to fall back on but really you earn very little as a PhD student and I had debt from both a master's and an undergrad that I was still paying off and credit card bills and everything else well that's a huge level of risk doesn't it to you and I think that's you know part of the point about entrepreneurship and um you know people that are from wealthier backgrounds can can find it easier because whilst they might not be given loads at least you know they've got the feeling of a safety net and all this sort of stuff so all this stuff matters you know it felt risky at the time and do a company but um yeah i i did and i suppose once you're in it once you've you've taken the plunge and it is really an important step to say i'm going to be you know a full-time employee of this of this company um once i took the plunge into running biorelate it was really then just a fight for survival for the first year or two, just trying to get going, trying to understand how things should work.

20:18I think some of the things that are really important when you throw yourself into running a business is actually changing your mentality completely towards running a company. And that's really, really hard. If you've been a PhD student, you're surrounded by your peers who are all PhD students, the kind of norms are very, very different when you're running a company. So I had to try and break out some of those, create a culture within my own life of entrepreneurship, which for me involved sitting in the business school and listening to lectures from business students, going to meetups and startup competitions and just surrounding myself with other entrepreneurs and other people that were in this kind of mindset and then you pick up their norms you pick up their behaviors and you start to see what good looks like you start to feel what um should be doing what you shouldn't be doing because there's lots of lessons to be learned when you're starting a company um so a lot of people talk about mentorship you know kind of in the early years of starting a business.

21:27I did have some mentors. I had a VC. I had a couple of entrepreneurs. I had, you know, some people in industry as well. I think all of those people are really, really important for just being sounding boards and having people to talk to and to just, yeah, stay sane, particularly as a solo founder as well. It's a completely different experience starting a company if you're doing it on your own versus having a co-founder because i think often co-founders they obviously have the the companionship um of you know being able to share the pain with somebody else doing it on your own for the first time it really is hard um particularly hard i wouldn't advise somebody to do that actually i'd probably call it one of my biggest mistakes is starting by relate on my own um if i'd have had somebody to share the load then i wouldn't have had to learn every position in the business to start with i could have just focused on you know being a ceo or cto or one of those you know so um yeah it was very very difficult and i think when you've got a finite amount of time and the things that you have to do are all very hard and you have to do them all very very well um just having somebody else share the load is is really key so i think a minimum having one other co-founder is important in starting a company um one of the advantages of being a solo founder is though that if you get past those early years and you've built a team you've built a really good senior team beneath you um that's where it becomes less of an issue because you've kind of got if you've built a very very strong senior leadership team that works with you they essentially act as your co-founders okay um so um yeah really really difficult um i could dig right into all the pain and misery of starting a company but um yeah you just have to throw yourself into it and um for now having mentored other people through the journey and just seeing the pain they go through and that and just the obvious mistakes that people make as well the only way to really learn is is just to do it I'm so glad that you started with mentality Dan because I I I see this a lot because I my side of the market I guess you know Jess and I just did our first angel investment um and so that sort of precede seed early you know early everything I guess part of the part of the market in health care and biology and health tech and biotech is obviously and naturally full of people that want to be in healthcare.

24:13And it's people that have often been through academia, like you've just said. And these worlds of, I mean, you literally call it commercializing research, don't you? There are tech transfer offices that help with this to commercialize your research because it's just not... That's not topics.

24:36Yeah. But yeah, those deals never seem to look good. But anyway, we could probably talk about that. But I think the mentality is absolutely right. I think it's so interesting that you placed yourself geographically, physically into the space of business students and where business was happening so that you could start to adopt those behaviours. I think that's really, really interesting. It's almost like method acting, isn't it? Like immersing yourself in the environment to start taking on those things. Even if you don't necessarily think them or feel them, you can start acting that way. And then eventually your actions will dictate your mindset and it will go the other way around.

25:19And I think that's a really smart thing to do. Acting learning, you learn a lot as well, right? So you're actually just listening into these lectures. you pick up um pick up knowledge that you wouldn't have picked up before because you spent all your time doing doing science right it's a very very different thing running a business um reading books as well um i read a lot of books in the early days um yeah good books as well not like these kind of self-help books but i think one i find it really really important to read kind of autobiographies from other entrepreneurs that have been successful and just hearing how they did it and you know being an engineer i often find like physics books and maths books also pretty interesting um yeah so reading is is a key skill as well i think i think you need to acquire enormous amounts of knowledge in order to do that job well so you have to be hungry for it and you have to seek it out and you can't expect to just sit in a room and build a company in your basement without um without really beginning to think about those things unless you're the cto and you've got okay i want to talk about that next bit then commercialization so turning this idea rudimentary idea a project and some publications with the fizer project turning that into an actual business so where does this idea start or actually i again i like the fact that what you're saying here there's no light bulb moment there's no initial bit where all of a sudden it was this thing and successful this was all a slow burn it was two months to even do the fires like i i like this it's real it's it's realistic and i think only once have i ever been on this podcast and heard someone say i was in the pub and had six pints and then on the way home thought of this thing that has a that story hasn't come out of it once um so that isn't a thing so how how did you then iterate this sort of nebulous thing that you've got here with this ideas and these publications and projects and at what point does this start forming into a business and actually let's zoom in on like the business model because that's what will determine this is a business right turning it into a product and then being able to sell it so how does that start to emerge in this almost just put yourself in the shoes of somebody that has just started a company hasn't run a company before has only a dwindling amount money and um and you know is fighting to to build a company my heart rate's gone up already i'm stressed your your number one goal to begin with is to get some funds into the company right through one means or another which gives most people two options a b either you you raise money or you um or you go out there and do some business you do some sales but if you don't have a product how do you do sales so um there's one thing you can do if you don't have a product and that's services so um you essentially do service work to generate some revenue that helps you build your products or your first products that then can help essentially keep the lights on while you're progressing through to then having enough to show to raise some money, which is pretty much what I did.

28:33So it was going out there, finding other companies that had similar issues to what Pfizer had. So they had a drug or they had a disease area that they were working on and then essentially building a database to then analyze that data to support um their improved decision making um so did this in the early days um but actually you know from companies all the all the way through to the big ones down to the the small ones um so yeah i went out there and did some business to begin with without having your products which um slowly helped me um understand how to run a company how to um you know how to do sales how to um do good customer service you know um essentially all the ins and outs of running a business um did it learn through doing service projects to begin with many of those were pretty successful and um and yeah that also helped me understand the product i wanted to build as well in a lot more detail so if you think building these databases on a one-off basis is not um is not the most scalable business model you can only really grow the business linearly if you um if you're building lots of things under a service contract as well there's no recurring revenue associated with that so most companies that scale and scale successfully do so via recurring revenue streams so they um in our sector it's um mainly a licensing model so you're essentially getting people to spend a certain amount of money for a period of time like a year um which then they will then hopefully renew at the end of the year if they continue to get the value that you initially signed them up for.

30:30So licensing is the holy grail, but you need a product builder license before you can do that. So yeah, we had really the first kind of demos for products before we did our first funding round. And then we used funding from investors to then build out the team to then productionalize these early versions of products so we went from being a service company in the early years um where we're like the technology we had to build was pretty difficult by the way it wasn't like uh you're just trying to put together like an e-commerce store or something like that this was like writing software to understand um literature um and structuring it to into a database which um you know the way that people were doing it then was through human beings reading through the stuff and doing it manually that was the only way you could do it well so um this is tricky and the technology available to do that at the time it wasn't like today nothing like it it needed to be um i mean you could innovate a hell of a lot back then because um you um you only had certain tools available to work with.

31:50So it's not like today where people can just go on to OpenAI, download an LLM, and usually you can get started with that and do some work that will essentially show that you can add value. But back then, you really had very little to start with. So you had to kind of build the whole thing from scratch, write all the code, train the models from scratch with your own training data, do all that kind of stuff. So it was pretty tricky building the technology back then. And yeah, I think once we had the demos and we showed what this could be, also showed that you could generate some revenue through service from decent biotechs and big pharma companies, that's going to gain the support of investors because they can then see that, A, this is a business model that sends for growth.

32:43B, you've proven that you can do this already. You've got some customers who are willing to pay for this and we can see from the products you've got. So yeah, and then you're on to the next suite of problems, which is building and scaling a company. GC is really, really difficult. It's about survival. It's about getting your first wins, finding the business model that, is really kind of it's early enough that then you can gain the support of an investor or investors to then turn that into a proper product market fit um afterwards um yeah so tough and you're you're bringing back a lot of painful oh man there's so much that i could talk about here um but i'll keep it brief so that we can move on from the painful memories the services first model i think is really interesting when you were actually introducing that little segment i thought you were just going to say yeah there's you know you can get a recurring revenue model and and you can build a product or you could just go to a VC and raise money you didn't say that you said or you can actually prove that you can bring money in yourself with a services model do a thing learn business learn these things and then go to a VC with you know revenue credibility and more knowledge of the product because you've been delivering that service in the area that your product is literally going to be it seems like obviously you know saying that after the fact it looks like an absolutely genius move i think a lot of that will have just been maybe you thinking of survival or maybe it was you and the genius and thinking it through because it's it's it never feels like a genius movie it may in hindsight yeah but it was mainly about survival but i just wonder if more companies could and should be doing that you know if i i see a lot of the business guru stuff online and and you know instagram tiktok all the rest of it and this does seem to be something that does tend to come up which is just learn to buy and sell something just do something that's relevant because to go to investors first and ask for money particularly as a first-time founder you're really gonna struggle and i think we've seen the world turn haven't we in terms of the availability of capital and it's yeah it does make a difference i mean it It does, yeah.

35:02The amount of money is available. You know, 2021 being the peak of that, it was much easier to raise money. Not to say, actually, I still think if you're a first-time founder and you're starting a company, it's pretty tricky to raise money at the beginning. I'm just, I'm seeing this from, you know, other people that, you know, I'm trying to help myself now today, go through similar pain. It's, yeah, no matter how good you look, you know, how successful your product might look as a potential, an investor is going to take quite a big punt on you if they're going to put money in at the very early stages.

35:40So yeah, the more that you can de-risk it from their own perspective, it's like putting yourself in the investor's shoes. Would you give money to a PhD student who said they can build a multi-million dollar company and it's the first time they've done it? Not many would, right? So you really do have to... prove that show them something that helps you from the crowds because uh yeah yeah yeah and my last my last question on this sort of early part of the journey before we move on to what biolo is up to now it's something that you alluded to actually which was um the sort of then versus now in terms of the technology and where the technology was i've written here on my notepad competitors question mark because as we sit here now in 2025 quite rightly as you said you could just go and open a and and do a you know a version of this in some way obviously not to the complexity that you're talking about particularly with the inferences and the understanding and the hypothesis generation and linking all together all the rest of there's lots obviously to it we'll come on to that but did you have any competitors at the time because as you tell the story in 2025 this this very much sounds like one of those ideas where you're like why hasn't this been done before yes you do have competitors so um and the kind of competitors i at the time were actually the companies that were were doing this manually so my whole pitch for disruption was that we were turning what was a manual process into an automated one um which we we've now certainly achieved that so the great thing about the kind of technology that's available today it means that you you can like literally turn very very manual kind of process like that into completely automated ones but that's still not easy that's still actually a very difficult process to get right um so my my point about you can come to open ai and you can um get started now with a model you don't have to train it yourself etc um it just means that the barrier to entry is lower but that you say you can get quite far quickly but then if you want to get to some real high grade stuff it's actually a pretty long tail of what you need to put in an innovation still um in order to get there and so i mean i'll give you i'll give you an example just a very very rudimentary one if you um let's say that you wanted to find i don't know drugs in published papers and you um you write some software that finds drugs in papers and you you write something that then finds papers that might have drugs and you do this for like 10 000 papers and you pull out the drug names from those papers um and then then your next challenge is right so let's try and scale that up to all of the literature so let's um put in like let's do 100 000 papers and the dark and order of magnitude again let's do a million papers and then let's do 10 million so very quickly you've gone up three four orders of magnitude in terms of scale of task just to put out drug names um and then you've got all the other classes of entities and you've got a map that's different ontologies then you've got to find the relationships between them then you've got to build and you've got to build interfaces then you've got to um understand like how this stuff actually can be used to to support target selection target validation etc you've you've probably got somewhere like you know a thousand plus important innovations that you've got to make before you've actually got a complete product that you can sell at the kind of level that companies like mine and others um are selling at so it's pretty pretty difficult so it's why it's why you need to have a niche usually when you start and um the way that we did it was we went out there and built small databases for people to start with as i say through the service work and then use that to gradually scale up and solve those problems so we're 11 years old so it's taken us quite a long time to just solve these problems and even today you know we're a team of people um dedicated towards doing this kind of thing there's so much more that we could do that we don't have time to do um because we're focused on trying to do the core things well i think it's funny isn't it how the world's moved on to a point where we all have access to open ai and we can mess around with it and to your point we can get a certain distance we're now very different to 2013 I think we're at a point where that technology can be understood by quite a lot of people it can't be built by anyone except you guys and your competitors that are doing it but from a place of being able to adopt this from place of being able to go to a person in a pharma company to get them to understand it to get them to introduce them to you know you to the right person in the company whatever it seems like the whole world the whole scientific world and the biotech world and the pharma world will understand this a lot more in terms of oh okay i get that because of how the world's moved on do you think that people talk a lot about timing don't they with entrepreneurship what's your timing been like then in terms of seeing this development through of large language models from where they were in 2013 to where they are in 2025 has have you been a beneficiary of timing or do you think it's counted against you in some way yeah it's a double-edged sword this one actually because um while that there's been much wider adoption of this type of technology is coming from a place of massive overhype so the expectations that people have about what it should do and how well it should do it are very very high to start with and and you're and you're also getting people that weren't experts in this stuff before who now proclaim to understand and be experts in it so um yeah it's kind of annoying but it's it's also a massive opportunity right because this technology does mean that we can now do some of the more difficult things that we would have considered impossible years ago.

42:16Like really, like, I mean, like technical examples, like we, before we could only find data within the sentences because the technology that we were working with just couldn't expand outside of that context window. It wasn't, the technology wasn't powerful enough. It's about it. No technology was, it just hadn't been invented yet. but then these really really big transform models that you know large language models we call them came along and suddenly we can then find data you know we can find relationships somebody's talking about something at the beginning of the paragraph and at the end of the paragraph and we can link those two things together and that little mini example means that we can then uncover a much much bigger pool of data which then improves all of the downstream hypothesis making that we do So me as a geek that's been trying to do that stuff for years, that's really exciting.

43:14Also, the kind of Vibe coding and agentic coding might be the nicer way to put it, is also really exciting because you can now build prototype applications really, really quickly without having to invest huge amounts of time and effort just to see if something makes sense or not. So you have these kind of throwaway apps that you can build in a matter of days just to see if it works, just to see if it's got legs. And then if it has got legs, then you can put it into production. I mean, this kind of stuff, this new way of working is really, really exciting. so yeah it's it's double-edged sword in that way it's also double-edged sword in the other way which is for I think most tech companies they that have existed prior to chat gbt a lot of your competitive advantage was based on your your knowledge of the sector your your um i'd say your your product market fit and everything that you've done um it it's a bit safer right because you've you've kind of you can defend a bit easier you can um essentially just double down on what you know and and continue building um a really rigorous technology from um your own niche that you've crafted um whereas i think a lot of the the kind of boundaries as i say have been dropped broken away and it's just opened up the um the sector to lots of people claiming that they can do the kind of things that you can do um and so that's that's a problem because you're you're having to stand out even more and you're having to make that make it much much clearer how you add value against all of these other supposed solutions that proclaim to do the kind of things that yours does but really they don't when when when it comes down to um to looking at the differences between them um so the climate that you're operating within is completely different product market fit is completely different as a result of this new technology um so you have to be moving with it and you have to be treating it um with um all the focus and effort um that if you don't um you will fall behind and you you will no longer have that product market fit so um yeah pros and cons completely here very relatable very relatable because even even for us like so much of that that you've just said you know we're a health tech and biotech specific communications agency whereas all of a sudden overnight every and you know that domain expertise was our differentiator whereas now overnight any generic agency can plug a question into chat gpt and get an answer and it will sound there or thereabouts and so really it's become quite interesting for us because our value for example is massively in things like this it's our distribution because actually very quietly we've been using these things like this podcast and health tech pigeon and all these things to kind of prove the point put our money where our mouth is and and and never really thinking of it as a media company never never considering ourselves as this publishing house that owns loads of publications and owns loads of distribution we haven't really considered it that way i tell you what we're really considering it that way now because actually you turn around five five six years later from building these audiences low and slow and gradually and all of a sudden they've got you know 10 20 000 people eyeballs watching this stuff and all of a sudden you realize hold on a moment this is not just a proof point that we can do what we say we can do actually we can just distribute the messages of people that are you know paying us for exactly that you know we're not having to rely on press and pr and and you know we're not at the behest of editors who can say yes or no whereas actually we can control a lot of this ourselves and i think there's a really interesting even path for us but um yeah so i get it man that the value changes it really it really has um but i want to talk about biorelate now and I want to talk about what the cool stuff is that you do now and maybe what you could do is walk us through like a really cool project that you've got going on now or a client that came to you and asked for something um and what your technology does because I want to geek out on this technology with you man because I think I think this is super cool from someone that's come from you know clinical medicine and you know written papers and been published and all this sort of stuff and had to do literature reviews and write dissertation and the rest of it like it's interesting to me this stuff and i love tech and so i think walk me through a pharma company coming to you or something similar and and what your tech has done and what the actual end of the day outcome is because let's not forget there's patients at the end of this that receive value as well yeah so what what we what we have is a platform that we call galactic ai so this is the this is our our kind of main product this is the the engine for which we do all of our data curation so it's by the way dan you definitely learned some marketing stuff before you named that so some point in the journey that you've talked about you said you started off not being very good at marketing naming stuff like you did turning around with galactic ai at the end of the day like clearly you picked up some stuff we trademarked that and uh yeah so we we built this back from Galactic AI and what it does is it ingests all of these different sources of text.

49:01So, passants, clinical trials, journal articles, grants, freeprints, basically anything that we can get access to without having to have a specific commercial license for to begin with. So, pull in all the data that we can get, right? And then we have this software, which essentially does really really difficult to perform data curation um so it turns unstructured text into a structured database um and the kind of data that we're trying to to map is like cause and effect um the really really valuable interesting data which help us understand how drugs work so um you know if like if you go through all of the things that a drug discovery company will want to understand about drug or drug target to begin with causal data is the main data type that will help you answer all those those difficult questions because it will tell you the mechanistic rationale it will potentially tell you the biomarkers it will tell you what other indications that your your drug could go for it will help you prioritize different targets against each other so it's a really really valuable data source and critically if you're pulling it out from literature it's grounded in evidence that is traceable and explainable.

50:19So both data scientists can use this stuff to come up with these really interesting hypotheses, and then they can pass it back to the biologist or pass it through to the biologists for them to be able to see exactly where these individual pieces of evidence have come from, so they know that they're not just dealing with machine learning scores or numbers that they don't really understand. So it's really, really important that this stuff is rooted in source. You have this traceability and you're essentially turning that into a structured data set. And then the way that we actually license it is through different products that we've created that sit on top of this data set.

51:01So one is called Galactic Web. And so really this is our main interface and it has lots of kind of search and insights driven functionality which is all guided towards kind of early drug discovery use cases. So we have like a biomarkers app in it, we have a mechanistic app in it and this kind of thing, and these will just help scientists try to answer complex questions. Then there's an API version of that, so that's for more of the data scientists, and then there's something that we call Galactic data itself which um is um literally just the data set so so companies can license any one of those different products and then they can use that in their research um in different ways so like if somebody's getting the data set typically they'll put that inside their own database with their own data and their own omics data they've generated and because our data is directional it's causal it will add all the directionality to their own data and help explain it in a way that they they've yet to understand thus far um so we've just literally finished version three of the platform this this month um and that's um we've now gone what we're calling fully lm native so we've trained our own large language models to be able to capture this kind of data from We're essentially training on 10 years worth of curated data that we've amassed ourselves from having an in-house data curation team that would read through publications and do this job manually so that we know what good looks like and the software knows what good looks like.

52:40And we've essentially trained and evaluated on that corpus. So it's now because we've trained really, really powerful NLMs on really accurate data. and you know this is deterministic so they're really fine-tuned to do that one particular job you know it's not like the kind of stuff that chat gpt is there for um it um it means that we're now capturing this stuff at kind of human levels of accuracy so for the first time in in all the years of running the company we have this full representation of all of the data that's ever been published in a graph. And it's just, it's really exciting to have that because once you have this data set, that just becomes the foundation for answering any number of questions that you might be interested in.

53:31But let me tell you about one of the really cool things that we're doing with this data set at the moment. And this actually came from the thinking from our new CSO that joined the company last year. That CSO was actually my supervisor at Pfizer. So I did actually get fired under eventually. So he's come on board and he's been transformative in helping bring some of the bigger insights from the bigger pharma companies into the business. so one of the ideas that he brought with him so ben actually joined from being the vp oncology data science astrazeneca so he was running a very big team there leading cross-discovery efforts to to develop oncology treatments one of the the areas that we brought with him was this was one of the ways that they will rate potential targets across lots of different biological criteria and you call it pts or probability of technical success and what this is is essentially eight different categories that if you go through them and try to understand each of those categories You upscore or downscore a target in terms of its potential to be a good target or a bad target.

54:58So one of those might be its mechanistic rationale. Another might be its biomarkability. And you just go through these different categories. And essentially what you want to do is understand as much about that potential target as possible before you then progress it further on into clinical trials. where it's potentially going to get a lot more costly um so it's both the act of finding good targets to work on but also the act of rooting out the bad ones and some people call this failing fast um so if you think that drug discovery with all the expenses that go into it one of the most expensive things to happen is if your drug fails really really deep into the clinical trials process so it fails in like a phase two you've incurred an enormous amount of cost up until that point and you've got really nothing to show for it because your drugs fail so one of the things that you want to do is essentially to root out those that are likely to fail as early as possible so fail fast so if you can pick up that there might be a safety complication for this drug from answering these kind of questions to begin with that might be one of the first things that you um you check off to make sure that a that isn't going to happen if it is going to happen um let's just drop it all together um so this idea of pts scoring it's really really rigorous you have all of these different categories um but each of these categories essentially to answer all of these questions well it requires um some pretty complicated analyses to be done and it might take something like six to 12 months to produce like a full report analyzing all the evidence and data you'll have around one of those targets but today that with the kind of agentic coding capabilities that are there you can essentially build one of these reports automatically within about 20 minutes if you code up the exact steps that need to be taken in order to perform these analyses right um so that that's literally literally what we did we we took all of the the pts criteria we took our brand spanking new data set that we've literally just released this month and then use that to help go through each of those questions systematically to rate potential targets as to whether they're good or bad.

57:42And so really it would have taken us, it would have taken us years probably to build an application like that from scratch without using these kind of agentic capabilities. And some of the features in it just wouldn't be possible so like the report writing element to it um so so we built this and and it has all of this this kind of traceability and provenance built into it so you can see all the way through to where every individual insights come from but also the steps that were taken to produce that analysis as well um so the real value of doing this apart from just having a nice report is that all of a sudden, instead of having to wait 6 to 12 months to do one of these PTS reports, and then that's it, you've just rated one of your potential targets.

58:33You can do this more systematically now across lots of targets. So you can actually use this as a target prioritization methodology in itself. So we think this is very, very new and novel. and this is something that we're um we're looking to bring in to get um you know get our customers to to be using at some point in the future so it's not released yet we're literally just playing around with the stuff ourselves but this has got me really excited because it's bringing together lots of different elements of the things we've been working on at the company for a large number of years we have this really really like high quality data set we have this very good understanding of the problems that drug discovery companies are facing.

59:16And then we have this amazing new agentic technology that's available. We're calling it Galactic Agents. We've built this kind of MCP-like access to our data that then gives us these capabilities to build these kind of tools. So it's all coming together in a very, very interesting way. And it's getting me excited. So I demoed this at a conference last week and seemed to go down quite well. um so that's one of the cool things we've been working on um yeah it's um it's an exciting time to be in this space it sounds awesome man like the question i have this might be a stupid question but one thing that you've talked about there is something that might have taken years can now take minutes and you know i might i might be slightly exaggerating there for effect but this point about real exponential or takeoff is upon us clearly now in what you've described in there somewhere lies the answers to you could argue every question that we have in healthcare how do you cure cancer how do you do this how do you do like all of that stuff is in that in that data somewhere if not right now it's coming in new research and new different bits and bobs so what i'm interested in is first of all is the value of what you've built infinite in some way given that research will always be added into this and this being the first port of call for people to ask questions on is the value what how do you even determine the value of what you've got what what what is creating more value in what you've done is it the actual application to the customer and figuring out the best like is that the area that you need to work on as a company to build value in the company itself so and is the value infinite like that's that's a broad clumsily asked question but my set the second part of this is where are the pinch points now what is stopping us i'll be glib for the sake of being glib what is stopping us asking the question of this database what do we need to do to cure cancer curing cancer i really don't like as a premise because it's kind of implies that cancer is just one disease that we could go out i know i see you coming from why what's stopping us from answering these really really grand questions so i still think you have to know how to analyze the data and what kind of sub-questions to ask next, right?

1:02:07And do you think research is really, it's a series of steps which we take in order to generate new insights, but that's driven by asking the right questions all the way through. And so something like PTS scoring is essentially all the right questions to ask in order to determine whether a particular drug target is the right target to proceed and that's based from years of experience of people doing research and seeing what works and what doesn't work right and what's yours cancer is actually quite difficult to find the right questions to ask in order to to give you the insights you might be seeking and i would imagine it would be very very different from one cancer to the next um but coming to the question of how do we add value and how do we know that we're adding value um actually pts scoring is a really good way to frame it like that because what you're what you're doing is you're essentially if you can help a company avoid taking a an enormous amount of cost on a drug target spending all the reception time and effort working on that target and then that target fails down the line you might save that company 300 million dollars from that one particular insight right um and you could even show companies um how this could have looked if they'd have used this approach so essentially you you do what called retrospective analyses where you take targets which have been previously worked on and then you show um you you kind of slice the data at particular time point so say you take out all the data from 2010 onwards so you've just got the view of what science was up until that point in time and then you do the same kind of analyses that with that data and show you what your um what your insights would have looked like back then um based on the technology you have today and through doing that you can then show companies for the targets that have failed whether what the pts scores would have looked like in that particularly as a proof point to them.

1:04:18So that's a really nice way of showing the validity of these types of approaches. And then you have to assume that those kinds of insights will translate to the targets that people are working on today. But it's very, very hard to get a full readout on value until stuff goes through the pipeline right the way to market or to failure. So, you know, That's the space we play in. Everybody has the same problem there. It's often why R &D in pharma companies is often pretty difficult to justify the expenditure on marketing, where you get an immediate return on revenue. You can't bank a return on revenue from investing in R &D.

1:05:08you can bank a return on marketing um in terms of sales so um so yeah r &d has always been tricky in terms of being able to um directly tie outcome in terms of revenue and market growth um compared to the amount of money that you you put in the last question that i have for you mate before i let you go um because you're you're on holiday as well which uh listeners should be very grateful that you've uh decided to do this today but um i'm gonna go around the houses a little bit with this so i i play i play a fair amount of chess right like i i only learned the game a few years ago um play a fair amount and in the world of chess there's this interesting moment where obviously machines became better at humans and machines started to infiltrate into chess and it change the game because people would start learning based on what Stockfish was doing.

1:06:08Stockfish being the best chess engine. And because people were learning with Stockfish, Stockfish would do these utterly random moves in the opening, for example, or really early in the middle game, the advantage of which would only then be seen very, very, very late in the game. And all of a sudden it became like, ah, okay, we can see why it's doing that. So fundamentally, and in loads of different scenarios it's obviously changed the game of chess so that humans have learned with it so this sort of symbiotic relationship between humans and machines have now changed the game of chess some argue for the better in terms of its quality some argue worse because it's not purist in the way that we play chess as human beings the way this relates to what uh you've talked about is that previously drug discovery was a very human endeavor the hypotheses for drug discovery you mentioned were from humans and we're moving to a point where unbiased hypotheses by machines are now being proposed things that we might not necessarily have thought of previously like those chess moves from before um with the machines making those chess moves things that we wouldn't have decided or even thought to have done as humans um with ai you know in alpha go as well ai will make strange new strategies and rules that humans have never thought of so you know in chess there's an argument isn't there like is this good or bad here it's it's impossible to really say it's bad for any way if these things can lead to such good things but my question is that do you think that we lose anything moving drug discovery from being human and artistic in some way and reflective of the human consciousness in some way and moving it purely to machines or is this fundamentally a good thing because now this is purely now being this is going to a direction of complete lack of bias complete lack of art and actually just objective reality and truth yes it's actually a very good question and um the way i'd approach it is by saying that um so what we have today in drug discovery with the expertise that's been accumulated within r &d companies um is we have some incredibly talented people that've built up this pattern recognition that that they can spot good from bad very very easily and they are you know the kind the most successful drug discovery still being done that's being driven by by human beings and particularly at kind of smaller biotech companies where um these people are just the out and out experts in that particular therapy area or that particular drug discovery technology and they're very very focused on doing a very good job of that and then pushing that through to being successful and actually if you compare that against purely ai driven drug discovery companies, those that are just focused on being AI first to develop any kind of drugs, they've actually been far less successful in the last few years.