Latest / The BioHub - by Avetix / Episode 250 - Dr. Garry Pairaudeau, Co-Founder CEO & Dr. Anthony Bradley, Co-Founder CSO of Dalton Tx

Transcript

Episode 250 - Dr. Garry Pairaudeau, Co-Founder CEO & Dr. Anthony Bradley, Co-Founder CSO of Dalton Tx · The BioHub - by Avetix

• 0:06
  Hi everyone and welcome to another episode of the Biohub
• 0:09
  podcast. Today I'm joined by Gary
• 0:12
  Paradol, Co founder and CEO of Dalton, TX alongside Doctor
• 0:17
  Anthony Bradley, Co founder and Chief Scientific Officer.
• 0:21
  After working together at Accentia, Gary Anthony Co
• 0:25
  founded Dalton, TX to help bring the latest advances to AI drug
• 0:29
  discovery. In today's episode, we'll
• 0:32
  discuss their journey, the opportunities and challenges of
• 0:35
  AI in the industry, and where they see the field heading next.
• 0:39
  Welcome both to the podcast. Thank you very much, Amber.
• 0:44
  Nice to be here. And just to start us off, I'd
• 0:47
  love to hear about both of your backgrounds, but Gary, why don't
• 0:51
  you start us off a little bit more about your background and
• 0:53
  what ultimately kind of led you into this AI enabled drug
• 0:57
  discovery field? Excellent.
• 1:00
  Certainly. So I'm, I'm a chemist by by
• 1:04
  training and Anthony actually likes to mock me that I'm the
• 1:07
  only person in the company who doesn't actually write any code.
• 1:10
  So I'm something of an impostor in this whole AI drug discovery
• 1:14
  area. But I think what I bring is
• 1:17
  probably 30 years, at least 30 years experience in drug
• 1:21
  discovery. I spent most of my life working
• 1:23
  for AstraZeneca and what got me into this, it's quite an
• 1:28
  interesting story. I think is just, it was it was
• 1:32
  pure luck that I happened to have a really smart
• 1:36
  computational chemist, a guy called Ola Angvist, who worked
• 1:39
  in my team at AstraZeneca. And I remember, I still remember
• 1:43
  really, really clearly, he worked in Gothenburg and I was
• 1:46
  over in Gothenburg. He said, Gary, Gary, come to my
• 1:48
  office, come and look at this. And he took me down to his
• 1:51
  office and he showed me basically just some molecular
• 1:54
  structures flash flashing up on the screen.
• 1:57
  And I'm kind of like, yeah, right, whatever, and what am I
• 2:01
  looking at? And he said, look, it's
• 2:03
  evolving. It's generating these structures
• 2:05
  and they're evolving to be histamine antagonists.
• 2:10
  And I looked at it and it took me a little while to get my head
• 2:12
  around what he was doing. I thought, wow, this is, this is
• 2:15
  quite incredible. And I mean, it's, it was by no
• 2:18
  means the first time people had applied those generative AI
• 2:21
  methods to drug discovery, but it's the first time I'd seen
• 2:24
  them. And Oler and his team built this
• 2:27
  out into, into what's now become, I think almost one of
• 2:30
  the industry's standard methods reinvent.
• 2:32
  It's on its fourth generation. And that was what brought me
• 2:35
  into the into the AI drug discovery area.
• 2:38
  And at that point I thought this is this is going to change
• 2:41
  things. And that was probably 10 years
• 2:44
  ago and clearly it has changed things.
• 2:47
  Amazing. And Anthony, what about your
• 2:49
  background? Yeah, So I'm, I'm also a
• 2:52
  chemist, but far less in the way I love than Gary.
• 2:55
  I I basically started writing code during during my my PhD.
• 3:00
  And I think that was, that was where I really kind of, I
• 3:04
  understood some of the problems in drug discovery and, and kind
• 3:07
  of how no one really talks about AI, but I think it was already
• 3:11
  kind of how AI and computation could help.
• 3:13
  So I, I did my PhD with Blacksmith Klein's computational
• 3:18
  chemistry team. And the basic remit of my PhD
• 3:21
  was we've got an absolute ton of data and we just don't, there
• 3:24
  was a couple of amazing chemists there, Darren Green and Ian
• 3:27
  Wall, who kind of just didn't feel that they were making
• 3:32
  enough from the data that they had available that they were
• 3:35
  making the kind of objective, data-driven decisions that they
• 3:39
  could be making. And that's what I spent the
• 3:41
  whole of my PhD doing was building computational tools
• 3:43
  that could just much more objectively use the available
• 3:47
  data. And, and then all the way
• 3:50
  through my kind of academic career after that.
• 3:53
  And then, and then my, my life at Accentia was just spent
• 3:57
  basically trying to build computational tools and
• 4:00
  computational teams that could use available data to make much
• 4:04
  more informed decisions at drug discovery.
• 4:05
  Now, now that that kind of is what people see as AI enabled
• 4:08
  drug discovery. Yeah, and as you've just touched
• 4:12
  on there, you all work together at X Centre before about
• 4:15
  founding Dalton, TX, what point did you realise there was an
• 4:19
  opportunity to really build something yourselves?
• 4:22
  For me, it was, it was actually a while after I left Exientia,
• 4:27
  so it wasn't something we were dreaming of while we were there.
• 4:31
  I think to be honest, I, I, I, we were far too busy building
• 4:35
  Exientia's capabilities to think about doing anything else.
• 4:38
  But it was after I left I, I was doing a mixture of some
• 4:42
  consulting and some gardening, to be honest.
• 4:45
  And, and it became clear that you kind of leave this verified
• 4:52
  bubble of a tech enabled company and go out in the sea into the
• 4:56
  wider world. And you realise that there are a
• 4:59
  lot of companies at various different stages of their
• 5:04
  understanding and adoption of AI.
• 5:07
  And I remember sitting actually in a restaurant bizarrely on
• 5:10
  Charlotte St. talking to Charlotte.
• 5:12
  And I remember the evening really well because for a
• 5:15
  particularly bad reason, I had my phone stolen, which is
• 5:18
  really, really annoying. But before it was stolen, we, we
• 5:21
  were chatting and we said, yeah, we've, we've got to do
• 5:24
  something. We, you know, she was saying,
• 5:26
  I'm developing all these capabilities in my group.
• 5:28
  We've got to get them out to people.
• 5:30
  We've, we've really got to do something.
• 5:32
  And then a few days later, I talked to, I talked to Adrian.
• 5:34
  I talked to Adrian Russell. And he coined it beautifully.
• 5:37
  He, he said to me on the phone, he said it would be remiss if we
• 5:41
  didn't do something, if we, if we didn't take the knowledge,
• 5:45
  the experience, the expertise that we had and, and use it to
• 5:49
  help people. And for me, that was the that
• 5:52
  was the tipping point. So I think there's kind of two,
• 5:57
  two parts to it. The first part was what I saw at
• 6:01
  Accentia kind of completely blew my mind because I, I on the
• 6:06
  first project I was on and it was part of an amazing project
• 6:10
  team. But we delivered a clinical
• 6:11
  candidate against PKC detail that's now with BMS in 10 months
• 6:15
  and a couple 100 compounds. And that is not normal.
• 6:19
  And it was probably the most bizarre first project to ever be
• 6:24
  on to, to deliver a clinical candidate that quickly.
• 6:28
  And, and so you could see and, and other people are showing
• 6:32
  that and more and more now as well.
• 6:34
  You could see that there was something clearly massively in
• 6:38
  AI for drug discovery and how it could completely transform the
• 6:41
  space. And it kind of was the solution
• 6:43
  to the the question that I'd I'd been asking during my PhD in my
• 6:46
  previous life. And then I think the second bit
• 6:51
  came after we left Acentia or around that time where what we
• 6:57
  could see was models understanding of the world
• 7:03
  around us with things like ChatGPT was growing massively
• 7:08
  and it was now possible to capture a bunch of the
• 7:11
  intangibles that we never were able to properly capture within
• 7:15
  a system in a proper what we could kind of now call it a
• 7:19
  world model of drug discovery. We could now see how we could
• 7:22
  capture those intangible things and that that's super exciting
• 7:26
  because because drug discovery is not just about predicting
• 7:28
  potency and then picking, picking the top molecules from
• 7:31
  it. It's really complicated and
• 7:33
  there's loads of like fuzziness and, and bits of it that are,
• 7:36
  are really hard to encapsulate in a really simplistic
• 7:39
  algorithm. And so that that got me super
• 7:42
  excited for us building Dalton because now we could find
• 7:46
  capture a lot of fuzziness within our amazing engineering
• 7:52
  platform and within those AI algorithms.
• 7:54
  Yeah. No, that's super interesting as
• 7:56
  well. And Gary, then 4 listeners
• 8:00
  hearing about Dalton, TX for the first time.
• 8:03
  Can you tell us more about the company, what you're building
• 8:06
  and that problem you're trying to solve?
• 8:09
  Yeah. And I mean, kind of building on,
• 8:12
  on what Anthony's just said, because actually that that
• 8:15
  really encapsulates the, the reason for setting Dalton up.
• 8:19
  I mean, as I said, we kind of looked at, we looked at the
• 8:22
  world and we really wanted to help.
• 8:23
  We wanted to bring that expertise that we had the
• 8:26
  learnings from XCNTR and a little bit from AstraZeneca and,
• 8:31
  and help people apply them so that they could discover drugs
• 8:33
  more effectively. Because let's be honest, I mean,
• 8:36
  creating medicines is probably one of the most important things
• 8:39
  we can do to help society. So we wanted to help people
• 8:42
  create better medicines and create them faster and more
• 8:46
  efficiently. And I think we we could dive a
• 8:50
  lot into the technical details, but at its simplest level, what
• 8:56
  we're trying to do in with Dalton is create a new operating
• 9:00
  system for drug discovery. We're trying to create a new way
• 9:03
  of people working. And there are hundreds, maybe
• 9:07
  thousands of tools, models, capabilities out there in the
• 9:11
  world. And they're, they're all
• 9:13
  important, but each one individually is, is only going
• 9:16
  to make a tiny impact. The impact comes when you change
• 9:20
  the way you work and you work in a different way.
• 9:23
  And that's kind of the root of what I think we, we learned at
• 9:26
  XCNT and what we built with Dalton now.
• 9:28
  And through I think the advances in agentic AI and the advances
• 9:33
  in being able to create these world models, as Anthony
• 9:36
  described, is a way of helping people change their workflow so
• 9:40
  that they use AI effectively and it actually delivers value, not
• 9:44
  just predictions. And that's what we're trying to
• 9:46
  build or not actually say, not trying to build.
• 9:49
  We've built it, people are using it, it's great.
• 9:51
  It works super well. No, that's great.
• 9:54
  And Anthony, from a scientific perspective, then, where do you
• 9:57
  think AI is really having the biggest impact on drug discovery
• 10:01
  today? So I think there's kind of again
• 10:05
  two parts to this answer this question.
• 10:08
  So one of the the first part is I don't think it's having enough
• 10:15
  of an impact for certain. And I think when you look across
• 10:18
  the space, it's not having that much impact yet on drug
• 10:22
  discovery. Drug discovery efficiency is
• 10:24
  still where it's been for a while.
• 10:27
  And yeah, I, I think the, the honest first part is I don't
• 10:31
  think it is for someone like me, it's not having anywhere near as
• 10:34
  much of an impact as I'd like to see.
• 10:36
  But then the, the slightly more positive answer where we can see
• 10:39
  it, where we can see it being used is it's very clearly when
• 10:45
  the companies that are employing it, it's reducing the time time
• 10:47
  to time to be able to discover a candidate.
• 10:50
  And I think that through a couple of different things.
• 10:53
  Firstly, you're able to much more efficiently and effectively
• 10:57
  learn over over project space and be you're, you're able to
• 11:03
  just reduce through a discipline process for using models.
• 11:06
  You're able to reduce the number of compounds that you make or
• 11:09
  the number of antibodies that you design that don't really
• 11:12
  either move the project forward or ask a meaningful question
• 11:16
  that allows you to build, build a better model over time.
• 11:19
  And so I, I think rather than, and where a lot of people maybe
• 11:24
  get excited rather than thinking about it as being, you're going
• 11:27
  to design your drug candidate in one molecule and that's all
• 11:30
  you're going to have to have to make that.
• 11:32
  That's not realistically what any of the scientific literature
• 11:35
  is saying at the moment. What?
• 11:36
  What we are seeing though, is that models are often above
• 11:42
  human performance levels at being able to predict a
• 11:45
  multitude of different properties and way, way above
• 11:48
  humans performance levels because humans can't think like
• 11:51
  this. At producing a set of compounds
• 11:53
  that helps the model to learn over time and helps to kind of
• 11:56
  accelerate learning and, and move things forward.
• 11:58
  And that that's, that's what's been proven by a bunch of
• 12:02
  different companies and a bunch of different academic labs as
• 12:04
  being, you know, demonstrably better than the previous
• 12:07
  process. Yeah, that's really interesting
• 12:10
  about it maybe not impacting the market as much right now.
• 12:14
  And I suppose, Gary, then there's a huge amount of
• 12:18
  excitement around AI at the moment.
• 12:20
  How would you distinguish between meaningful progress and
• 12:24
  all of the hype that's out there right now?
• 12:28
  Yeah. That's, I saw that question on
• 12:31
  your sheet that you sent to us and I was sort of dreading it
• 12:34
  because it's kind of, and sometimes hype can turn into
• 12:41
  meaningful progress. And actually, I doubt meaningful
• 12:44
  progress can ever turn into hype.
• 12:46
  But I think the, I think the, the key to this is, is in the
• 12:50
  word meaningful. And what I've, what I both love
• 12:54
  and hate about my whole time spent in drug discovery is it's
• 12:59
  a real leveller. At the end of the day, you have
• 13:03
  to do experiments, and those are real physical experiments.
• 13:08
  And a question I've, I've been asked quite a lot is what's the
• 13:14
  difference between an AI drug going into the clinic or an AI
• 13:19
  design drug going into the clinic and a more traditionally
• 13:22
  designed drug going into the clinic?
• 13:24
  And the answer is nothing, absolutely nothing.
• 13:27
  In terms of the actual molecule and the actual tablet, there is
• 13:31
  no difference at all. It still has to pass exactly the
• 13:34
  same exacting standards by the FDA or whatever agency is
• 13:38
  approving it and it has to satisfy the same safety
• 13:42
  standards etcetera. So at the end of the day,
• 13:47
  whatever predictions you're using, whatever computational
• 13:50
  methods you're using, they're going to be levelled by drug
• 13:53
  discovery because you're going to have to create a molecule and
• 13:57
  that molecule is going to have to have an effect ultimately in
• 14:00
  patients, but along the way in biochemical assays and cellular
• 14:04
  assays in in animals. So, so I think meaningful
• 14:09
  progress is doing something that actually has a, an impact there
• 14:14
  and actually has an impact on that process.
• 14:16
  And by impact, I mean really, really simply enables you to do
• 14:20
  something you couldn't do before or enables you to do something
• 14:23
  more efficiently. So faster and cheaper.
• 14:26
  And as you know, as Anthony really nicely described, what
• 14:30
  we're seeing from some of the leading AI drug discovery
• 14:33
  companies is, is definitely faster and cheaper.
• 14:37
  Without a shadow of a doubt. I, I think I'm sure we will see
• 14:41
  evidence of things you couldn't have done by any other means as
• 14:45
  well as we move forward. But for me, the, the, ultimately
• 14:49
  the proof of the pudding is always going to be experimental
• 14:51
  validation. Yeah.
• 14:53
  Definitely. And I suppose we've announced
• 14:55
  partnerships already across both the small molecule space and the
• 14:59
  antibody space. What have those collaborations
• 15:02
  taught you about what the industry is looking for from AI?
• 15:07
  Yeah, I'm going to try and think of a different answer to the
• 15:09
  last one because the answer is really the same.
• 15:12
  That's what people are looking for is impact.
• 15:15
  And I mean what what you know, we've, we've, we collaborate
• 15:19
  intentionally with biotechs and with CR OS because that's where
• 15:24
  an awful lot of the drug discovery in the world happens
• 15:27
  at the moment. And we, we set the company up
• 15:30
  because we wanted to help the world create that seems more
• 15:33
  effectively. And we want to help real world
• 15:35
  drug discovery. And that goes on in a lot of
• 15:38
  biotechs and a lot of CR OS and they're wholly focused on
• 15:43
  impact. You know, if you're a biotech
• 15:46
  where you care about is, can you advance your molecule before
• 15:49
  your money runs out? So they want to go faster.
• 15:52
  That's, that's, that's what they want.
• 15:54
  If you're ACRO and you care, you're, you're ruthless about
• 15:59
  effectiveness because that's what you're paid for.
• 16:02
  And so it's a, it's a tough, but I think really enjoyable
• 16:06
  environment to be working in because they actually value the
• 16:09
  things we value, which are, you know, meaningful improvements in
• 16:14
  performance. Anthony.
• 16:17
  Anthony may want to add something on this.
• 16:19
  He also loves this subject. Yeah, yeah.
• 16:23
  I mean, I, I think that's the real beauty of working with
• 16:25
  real, real partners, real businesses that have, you know,
• 16:30
  very tangible, often really reasonably short term goals that
• 16:32
  they want to achieve. And it massively keeps you
• 16:35
  honest and keeps you focused on not building shiny algorithms
• 16:39
  for the sake of building shiny algorithms, but for doing the
• 16:42
  things that are needed to help projects deliver more, more
• 16:45
  quickly. And I, I think what we've built
• 16:48
  and what we're using is, is able to do that, which is, is
• 16:51
  amazingly exciting. Yeah.
• 16:54
  And I think that comes on to kind of the next question that I
• 16:56
  was going to ask pretty well. You know, you're bringing
• 16:59
  together that software engineering, AI and drug
• 17:02
  discovery expertise. Why is it really important to
• 17:05
  combine those disciplines when building technology that
• 17:08
  scientists you know can actually trust and use?
• 17:12
  Yeah, because you absolutely. And this is the thing I think
• 17:17
  both Gary and I have learnt the hard way by not doing it like
• 17:22
  this. But you, you learn that you
• 17:25
  absolutely need all three. And I mean everything within AI
• 17:31
  has shown this. Most of the major advances in AI
• 17:35
  have been driven at the confluence of AI and
• 17:38
  engineering. People being able to scale
• 17:40
  compute systems, being able to design and develop algorithms
• 17:44
  that actually fit with the hardware that you have available
• 17:47
  and the way that you engineer software.
• 17:49
  And if you don't do that, you create so much technical debt so
• 17:54
  quickly that you you, you, you, you can't move forward.
• 17:58
  You get stuck in treacle and and you have to be able to build
• 18:02
  systems that fit with an engineering paradigm.
• 18:05
  And then at the other interface of AI and drug discovery, again,
• 18:09
  it comes down to, you know, algorithms that sound, you know,
• 18:14
  you, you can come up with really fancy, amazing sounding
• 18:16
  algorithms that look conceptually like they should
• 18:20
  help. But without a real understanding
• 18:23
  of what the problems are in drug discovery, it is impossible to
• 18:26
  build, I think, useful AI products.
• 18:28
  And again, that's what you've seen time and time again, the
• 18:31
  people who've moved forward have had some degree of the
• 18:35
  compliments. But we, we think we've built a
• 18:38
  pretty exceptional team that leads at the ability to speak
• 18:42
  those 3 languages fluently and be able to corral a team that
• 18:46
  that can deliver on that. Yeah.
• 18:50
  Gary, do you have anything to add on to that point as well?
• 18:54
  I I don't really it was a great it was a great answer, wasn't
• 18:58
  it? As we'd expect.
• 19:00
  I mean, yeah, this is this is what's important.
• 19:02
  And I think, you know, we, we have a couple of other Co
• 19:05
  founders in the company. Charlotte Dean is obviously a,
• 19:08
  you know, world leader in the the scientific discipline and,
• 19:12
  and Adrian Russell, who built Astrazeneca's AI design
• 19:15
  capabilities and then built Exientias.
• 19:17
  And, you know, so we have sort of really deep, really deep
• 19:21
  engineering expertise as well on the team, which as Anthony says,
• 19:24
  is, yeah, it's, it's super, super critical to bring all
• 19:28
  those three things together. Yeah, and I think moving on to
• 19:31
  the team then you know you've grown the company to more than
• 19:34
  25 people in a relatively short period of time.
• 19:38
  What have you learnt about building both the team and the
• 19:40
  culture as first time founders? Yeah, that's, that's
• 19:45
  interesting. I mean, you, I, I don't know
• 19:47
  whether the video is going out, but you can probably tell I'm
• 19:49
  recently old. So it's, it's not my first time
• 19:52
  building a team, but it is my first time running a company.
• 19:55
  So, so that has been a hugely interesting experience.
• 19:59
  And it always strikes me that, you know, we talk about AI, we
• 20:02
  talk about technology, we talk about engineering, but it always
• 20:06
  comes down to people. It it always comes down to can
• 20:10
  you, you know? Can you get a group of
• 20:12
  exceptional people and get them to work together?
• 20:16
  And I think, I mean, we've done the amazing job it it.
• 20:22
  I still get goosebumps on the back of my neck when I walk into
• 20:26
  the office. Tuesdays and Wednesdays, the
• 20:28
  whole team's in the office and it's rammed.
• 20:31
  And I, I, I walk in and it's, it's just a buzz.
• 20:36
  And it's a buzz of people talking about science and
• 20:40
  software engineering, what they've just built and how
• 20:42
  they're going to apply it and what the latest data from our
• 20:45
  partners says, etcetera. And it's, it's such an exciting
• 20:49
  atmosphere. And I think I mean, actually,
• 20:53
  you know, bring Anthony in as well.
• 20:55
  I mean, he's done. Anthony and Adrian have done all
• 20:57
  the heavy lifting on actually hiring the team and, and
• 21:00
  bringing them on board. But in in some ways the culture
• 21:04
  makes itself. We've got a really important
• 21:07
  problem. We've got fascinating solutions
• 21:11
  to it. We've brought exceptional people
• 21:12
  together and, and, and particularly with a small team,
• 21:16
  I actually think just don't mess with it too much.
• 21:18
  Just don't get in their way. Let them get on with things and
• 21:22
  they'll do amazing things and they are doing amazing things.
• 21:26
  Yeah, Anthony, anything to add on that, building that team out?
• 21:32
  No, I mean, I, I, I just feel amazingly lucky that we've been
• 21:36
  able to, to, that we continue to be able to attract such amazing
• 21:40
  people. And, and obviously I think being
• 21:43
  in London, being in King's Cross is super helpful.
• 21:46
  And and we, we kind of made the decision reasonably early on
• 21:49
  that we wanted to be office based and I feel like that was
• 21:53
  the right decision. Because.
• 21:55
  We have a really nice togetherness within the group
• 21:58
  and within the team and, and it those kind of, I remember
• 22:02
  someone ages ago and Nichols kind of talking about the
• 22:06
  culture open eye in the early days and they, they've talked
• 22:09
  about coming together to think and then going away to work.
• 22:13
  And I think for computational work that works really well.
• 22:16
  But you're not always in each other's pockets all the time.
• 22:19
  But you come together and you have two days in person and
• 22:23
  working together on hard problems and then you can go
• 22:26
  away and and build your solution.
• 22:28
  Yeah, I think it's so hard to build a really great culture as
• 22:31
  well. So it sounds like whatever
• 22:32
  you're doing with building that team is is in the right
• 22:35
  direction and. I think Amber, can I, sorry, can
• 22:39
  I have one other thing. What I really like is I, I, we
• 22:43
  have a group of people that just want to help each other and I
• 22:46
  think that's really important in a, you know, the one thing I've
• 22:49
  learnt about being in a start up and being in a small company is
• 22:54
  you kind of have to be prepared to do everything.
• 22:56
  And, you know, you have to take the bins out some days as well
• 23:00
  as think strategically about where we're going.
• 23:03
  And I noticed we Monday mornings, we have a kind of all
• 23:06
  hands meeting and you just get this vibe that if anybody's got
• 23:11
  a problem, there'll be a bunch of people that will weigh in,
• 23:14
  you know, oh, I've got some expertise in doing this
• 23:16
  particular thing from five years ago.
• 23:18
  I can help you or that. And, and it's just, it's so nice
• 23:22
  to see. And yeah, I think it just
• 23:25
  creates a great, great environment.
• 23:28
  Yeah, I love that. And thinking about the market
• 23:31
  then, if there would be or if there is one misconception you'd
• 23:35
  like to challenge about AI and drug discovery, what would it be
• 23:39
  as opposed to both of you? But maybe Anthony can start us
• 23:41
  off. Can I have two?
• 23:45
  You can have two. Yeah.
• 23:48
  So I, I, I think the first one is around the and it's sort of
• 23:53
  there's a sort of a scientific one and then there's a, there's
• 23:55
  a, there's a structural 1. So the, the scientific one is
• 23:59
  the, the model quality is the most important thing.
• 24:02
  And there's this misconception, I think that focusing in on
• 24:07
  model quality and benchmark is not unimportant, but it's not
• 24:10
  the most important thing. It is how you apply those models
• 24:14
  is way, way more important than trying to work out if you've got
• 24:17
  a slightly better R-squared or a slightly better spin and row
• 24:20
  than than another person. And, and I, I, I think it is,
• 24:26
  can be slightly limiting in the way that we then apply and, and
• 24:30
  actually genuinely deliver impact with, with things.
• 24:33
  And then the second thing is this kind of misconception that
• 24:38
  if AI drug discovery has never worked until we get a compound
• 24:44
  that's being, you know, marketed as a marketed fully approved
• 24:48
  drug, that's not true. You can still be delivering
• 24:52
  molecules into phase one and phase two at much greater
• 24:54
  efficiency and still and be demonstrating that AI drug
• 24:57
  discovery is effective because there's a whole load of
• 25:00
  different processes in drug discovery and there's a whole
• 25:02
  load of things that can go wrong that have, you know, as Gary
• 25:05
  said, nothing to do with the algorithms.
• 25:07
  By the time that they do. And, and I, I kind of think
• 25:09
  that, yeah, that conception is again holding back the industry
• 25:14
  slightly because it, it stops us realising what's already
• 25:18
  actually happening. Yeah, that's a great insight,
• 25:21
  especially around that. And Gary, what about you?
• 25:25
  Well, Anthony Hogg, the 1st 2 of mine.
• 25:28
  So I think I I just I'll say something really quickly, which
• 25:32
  is it's not all hype. So I think, you know, the, the
• 25:38
  almost the, the reverse of the hype is true that there is some
• 25:42
  amazing things being done out there using computational
• 25:46
  methods, using AI to really advance drug discoveries.
• 25:49
  So I think, you know, shouldn't believe everything, but there
• 25:55
  actually is a kernel of really valuable stuff happening and
• 25:59
  people can make the best of it. Yeah.
• 26:01
  And I think that brings us on to the next question Really.
• 26:03
  Well then, so looking ahead over the next five years or five to
• 26:07
  10 years, what excites you most about the future of AI enabled
• 26:11
  discovery? Oh, great.
• 26:14
  OK. But we so we've we've just
• 26:21
  started to think about kind of moving beyond individual
• 26:27
  predictions and into creating this idea of world models of
• 26:32
  drug discovery where you capture institutional learning, you
• 26:37
  capture reasoning, you capture how people are thinking
• 26:41
  strategically about solving problems, not just a prediction
• 26:45
  tells you where you are at the moment.
• 26:47
  A strategy tells you where you're going to go and where
• 26:50
  you're going to go is much, much more important.
• 26:52
  So I'm really excited about seeing how that shakes down and
• 26:58
  how that starting to really harness the power of collective
• 27:03
  thinking and collective readers name to solve much more
• 27:07
  challenging problems. So it stops it's you know,
• 27:10
  efficiency is really important, but what I'm really interested
• 27:14
  in is innovation. Can we create medicines that
• 27:17
  that couldn't have existed before?
• 27:20
  And actually, can we do that by capturing collective thinking?
• 27:24
  Yeah, And Anthony, would yours be the same or something else?
• 27:28
  I mean, obviously I completely agree, but I also think what I'm
• 27:32
  excited about is we kind of in the, the what I think the era of
• 27:36
  impact for AI drug discovery, we are now in the place where we're
• 27:41
  going. So I have a slide in one of my
• 27:43
  talks that kind of shows the, the, there's this exponential
• 27:48
  growth in, in the number of AI discovered molecules and
• 27:52
  targets, but it's still only a tiny fraction of the overall
• 27:57
  supply of AI of overall discovered molecules and
• 28:00
  targets. But because it's exponential and
• 28:04
  I believe it will continue to be exponential, that will suddenly
• 28:07
  be by the end of the decade, 4050% of all, all cargoes and
• 28:11
  molecules are being AI discovered.
• 28:13
  And that's super, super exciting because all of the promise that
• 28:16
  we've been seeing over the past 10 years will now be converted
• 28:19
  into actual impact. Yeah, it's a really exciting
• 28:23
  time as well. And just to round us off for the
• 28:26
  end of the podcast, then something we ask all the guests
• 28:28
  that join us and if you could take any innovator or scientist,
• 28:32
  past or present for dinner, who would it be and why?
• 28:37
  Maybe Anthony can start us off. I can start.
• 28:40
  I had to think about this. So I, I go for Dorothy Hodgkin
• 28:45
  in part because part of my PhD was doing structural biology and
• 28:50
  we've some people who are loosely aligned or in the kind
• 28:53
  of line of succession from, from Dorothy Hodgkin down.
• 28:57
  But she also just sounds like obviously like supremely
• 29:00
  intelligent, capable person, but also super fascinating as well
• 29:05
  in terms of her views on the world and, and, and, and how the
• 29:08
  world should be run and, and equity and equality.
• 29:13
  And yeah. And just, I, I don't know, I, I,
• 29:17
  I found X-ray crystallography fascinating when I was doing it
• 29:20
  during my PhD and just how it's actually really, really nice
• 29:25
  example of something that is so complicated and yet a complete
• 29:30
  Muppet like me can write something on a command line on a
• 29:33
  computer and it just works. And it does this complicated
• 29:36
  mathematics that I do not understand at all and is able to
• 29:39
  process this data that doesn't look like an structure at all.
• 29:43
  So it kind of be super interesting to talk to one of
• 29:45
  The Pioneers of turning it from this really complicated thing
• 29:50
  into basically an engineering problem.
• 29:52
  Yeah, that's a really interesting one.
• 29:53
  And what about you, Gary? I'm so pleased Anthony gave an
• 29:58
  intellectual answer to this question.
• 30:02
  So the first thing that came to my head is I wanted to take
• 30:05
  Keston Blumenthal to dinner for two reasons.
• 30:09
  Combines 2 things that I really love, chemistry and cooking.
• 30:14
  And we're going to get a bloody good dinner, aren't we?
• 30:17
  Indeed, that's a great answer actually from from both of those
• 30:21
  points. Well, thank you so much, both of
• 30:24
  you for joining us on the Biohop podcast today.
• 30:26
  It's been really great to sit down and talk all things AI.