Latest / Making Medicine / AI in Drug Discovery: What Works, What’s Hype & What Comes Next

Transcript

AI in Drug Discovery: What Works, What’s Hype & What Comes Next · Making Medicine

0:00
Speaker 1: At Making Medicine, we're reporting on the headlines, not making
0:03
Speaker 1: medical recommendations. For personal health questions, always ask your doctor.
0:08
Speaker 1: Nothing in this episode constitutes investment, financial, or legal advice.
0:13
Speaker 1: Consult your own advisors before making decisions.
0:16
Speaker 2: Clinical trial design, I think AI is going to be
0:18
Speaker 2: very powerful, but you have to be very deliberate about
0:22
Speaker 2: asking your question, and you have to make sure that
0:24
Speaker 2: the Context underneath your feet isn't sifting unwittingly, right? That's
0:28
Speaker 2: where AI becomes very bad. If the contextual rules change
0:32
Speaker 2: and you don't know they've changed, then AI is no
0:35
Speaker 2: better than guessing.
0:41
Speaker 1: Welcome back to the Making Medicine podcast, where science, policy,
0:45
Speaker 1: and innovation meet. I'm your host, John Stanford. Today, we're
0:49
Speaker 1: welcoming back a friend of the show, an Incubate Advisory
0:52
Speaker 1: Board member, Dr.
0:53
Speaker 3: Tom Kaiser.
0:55
Speaker 1: Tom is the co-founder and chief scientific officer of Avicenna Biosciences,
0:59
Speaker 1: which uses machine learning to speed up the last mile
1:02
Speaker 1: of small molecule drug discovery.
1:04
Speaker 3: He's also now an operating partner.
1:06
Speaker 1: At DCVC Bio, the deep tech life sciences venture firm
1:10
Speaker 1: that led Avicenna's seed round. Before Avicenna, he led the
1:14
Speaker 1: antivirals group in Emory's Leota Research Group, whose work has
1:18
Speaker 1: produced more than 20 FDA-approved therapies. He's also been one
1:21
Speaker 1: of the more credible voices cutting through all the AI
1:24
Speaker 1: hype most recently in a must-read Biospace op-ed, arguing that
1:29
Speaker 1: AI will only revolutionize drug discovery and development if we
1:32
Speaker 1: play to its strengths. Today, we're catching up with him
1:35
Speaker 1: on Avicenna's pipeline, what brought him to DCVC Bio, how
1:39
Speaker 1: he sees China reshaping the competitive landscape, why the pill
1:43
Speaker 1: penalty is still front of mind for companies like his,
1:46
Speaker 1: and he gives an absolute masterclass on how AI can
1:50
Speaker 1: and cannot reshape biotech. With that, let's get into the show. Tom,
1:56
Speaker 1: welcome back to the show.
1:57
Speaker 2: That's great to be here, John. Thanks for having me.
1:59
Speaker 1: Well, our listeners who have heard you before, and we've
2:02
Speaker 1: had you on to be a guru on all things biotech,
2:05
Speaker 1: all things AI, they know you as Avicenna, but excitingly,
2:09
Speaker 1: you're adding more hats since you were last on the show.
2:13
Speaker 1: You've taken on not just the running of the company
2:16
Speaker 1: and leading a company, but also being an operating partner
2:19
Speaker 1: at DC VC bio, um, maybe jumping off there, what
2:24
Speaker 1: made you take that leap and what are, what's different
2:27
Speaker 1: now that you also wear an investor hat?
2:30
Speaker 2: Yeah, well, it's, um, making that leap was really something
2:33
Speaker 2: that was offered to me. Um, so John Hamer called
2:37
Speaker 2: me up one day and he said, Hey, uh, do
2:39
Speaker 2: you want to be an operating partner in our firm?
2:41
Speaker 2: And I said, uh, sure. What does that mean? So
2:46
Speaker 2: it wasn't an area that was really on my horizon, but, um, Yeah, it's,
2:51
Speaker 2: it's a, it's a real privilege to join an exemplary team. The,
2:54
Speaker 2: the DCVC, DCVC bio community is just, uh, it's, it's
2:59
Speaker 2: chock full of really talented and nice people. It's a
3:02
Speaker 2: rare combination, uh, to find, but, um, they're the institutional
3:07
Speaker 2: investor in Avicenna. And so I think they really wanted
3:10
Speaker 2: to bring some medicinal chemistry expertise and some machine learning
3:13
Speaker 2: know-how into the firm to help evaluate all these exciting opportunities.
3:18
Speaker 2: in emerging areas like molecular glues, is this really something
3:21
Speaker 2: that we should invest in? And then also the rise
3:24
Speaker 2: of the frontier model, what does this mean? How should
3:26
Speaker 2: we think about it? So how has changed what I
3:32
Speaker 2: do when I think about Avicenna and just biotech in general,
3:36
Speaker 2: it's made it a tighter package in terms of presenting
3:40
Speaker 2: the opportunity, the upside, where we're going, what's really the
3:44
Speaker 2: innovation that we have to offer. But It's amazing to
3:47
Speaker 2: see the vanguard of things that are just rolling out
3:51
Speaker 2: of university. You get to kind of get a sneak
3:53
Speaker 2: peek before they go non-stealth or where are people thinking
3:58
Speaker 2: about what is going to be the new big thing.
4:01
Speaker 2: So it's been really cool to see. And then also
4:04
Speaker 2: coming in and trying to help companies tighten up their
4:08
Speaker 2: IND packages and really select candidates. That's been fun. Just
4:13
Speaker 2: good old-fashioned MedKim. Nothing fancy. Are we on the right
4:17
Speaker 2: path for designing a medicine? You know, the old bread
4:19
Speaker 2: and butter of what I was trained in before all
4:22
Speaker 2: this AI stuff. So it's been great. I've really, really
4:26
Speaker 2: just enjoyed it. It's been a lot of fun, a
4:29
Speaker 2: lot of breadth, a lot of new stuff to see
4:31
Speaker 2: and think about.
4:32
Speaker 1: Is it changing anything? Does it change how you look
4:35
Speaker 1: in the mirror from an Avicenna standpoint? Like, has your
4:39
Speaker 1: perspective shifted of, oh, now I kind of get why
4:42
Speaker 1: the investment community is asking this questions or they like
4:45
Speaker 1: data presented this way?
4:47
Speaker 3: Or is it all still pretty rote?
4:50
Speaker 2: Yeah, we were really fortunate, Peter and I were, that
4:54
Speaker 2: DCVC put so much into us. One of the entrepreneurs
4:58
Speaker 2: in residence from DCVC Bio actually came into Avicenna. So
5:02
Speaker 2: we had a very good idea early on of how
5:05
Speaker 2: to present ourselves and how we need to package and
5:08
Speaker 2: present data. And that's really a testament to Chris Meldrum.
5:13
Speaker 2: And I think we've certainly thought more carefully about that.
5:18
Speaker 2: As you get close to the clinic, the bar definitely
5:20
Speaker 2: goes up quite a bit in terms of what investors
5:22
Speaker 2: demand and expect. But I think it's, if anything, it's
5:27
Speaker 2: made It's made it quite clear that what we're doing
5:31
Speaker 2: in terms of focusing exclusively on candidate design is still
5:34
Speaker 2: very rare. A lot of people want to do new biology,
5:38
Speaker 2: new chemistry, and there's a lot of risk with that.
5:40
Speaker 2: Whereas we want to do drug design, like exclusively clinical
5:44
Speaker 2: candidate design. So it's been nice to really understand that
5:49
Speaker 2: at a very high level and be able to pitch
5:52
Speaker 2: that story a little bit more carefully and clearly.
5:54
Speaker 1: Well, and the story is only getting better. So let's
5:56
Speaker 1: dig a little deeper on Avicenna. My understanding is some
6:00
Speaker 1: congratulations are in order as you've gotten some good news
6:04
Speaker 1: out of the FDA on your pre-IND for first in
6:08
Speaker 1: human trials around your ALS drug. Tell us a little
6:12
Speaker 1: bit more about that and the good things happening at Avicenna.
6:15
Speaker 2: Yeah. Thanks a bunch. I mean, it's always exciting when
6:18
Speaker 2: you interact with regulators and they come back and say, yep,
6:21
Speaker 2: you're on the right track, right?
6:23
Speaker 3: That's a good sign.
6:24
Speaker 2: That's what you hoped to hear. So we just got
6:28
Speaker 2: our pre-IND response letter back, and it was very good. Basically,
6:34
Speaker 2: that's when you lay out the last things that you're
6:36
Speaker 2: going to do preclinically, your GLP talks, and then your
6:39
Speaker 2: prototypical studies for your first in human. And you basically
6:42
Speaker 2: ask the FDA, hey, this is the tox package that
6:46
Speaker 2: we think we need in order to have a safety
6:48
Speaker 2: profile to go first in human. What are the things
6:51
Speaker 2: that we're missing, if any? And we got a very
6:54
Speaker 2: clear answer that we're on the right track, that this
6:56
Speaker 2: is a good package. So it's always exciting to, I mean,
7:01
Speaker 2: we literally designed the molecule, we picked the target, we
7:04
Speaker 2: picked the indication, we did everything in Avicenna. So five
7:08
Speaker 2: years from literally turning the key to pre-IND is amazing.
7:12
Speaker 2: And it's really exciting that we're going to go into
7:15
Speaker 2: an area of huge need like ALS and hopefully have
7:19
Speaker 2: a disease-modifying drug there. So it's a testament to the
7:23
Speaker 2: power of AI in drug design when you use AI correctly, right?
7:27
Speaker 2: And we can talk about that a little bit, but
7:29
Speaker 2: it's a very exciting time. It's amazing that we've had
7:33
Speaker 2: such a wonderful group of people that are very skilled
7:37
Speaker 2: at what they do. It's a tiny number of people
7:39
Speaker 2: that are really pushing us forward. And normally you have
7:42
Speaker 2: We were able to do this with $ 25 million. There
7:45
Speaker 2: are companies that have raised $ 250 million, half a billion dollars,
7:49
Speaker 2: and they're further behind than we are. So really exciting
7:53
Speaker 2: and proof of the fact that we're doing something neat.
7:56
Speaker 1: Well, and I want to stay there because, yeah, you
8:00
Speaker 1: are sort of doing this on a shoestring budget, and
8:03
Speaker 1: maybe that's part of the story around AI because, as
8:05
Speaker 1: you've said– You have 20 some odd million put into this.
8:10
Speaker 1: And not only do you have your rock inhibitors, you
8:12
Speaker 1: have your CSF1R program.
8:15
Speaker 3: So you've got.
8:16
Speaker 1: These things coming along. What allowed you to do this
8:21
Speaker 1: at such an efficient scale? Were you more disciplined? Did
8:24
Speaker 1: you watch cash flow? Did you feel like your belts
8:28
Speaker 1: were tight or were you just being more efficient leveraging
8:32
Speaker 1: AI and ML?
8:34
Speaker 2: Yeah, well, it's a combination of both. We were very
8:38
Speaker 2: careful to build a community that knew, here's a key
8:43
Speaker 2: point in the history of the company. This is going
8:45
Speaker 2: to be a make or break it. We can't really
8:47
Speaker 2: pay these exorbitant salaries. We've got to be very lean
8:50
Speaker 2: on how we investigate programs. So we really adopted a
8:54
Speaker 2: fast to fail where we make some compounds, we test
8:57
Speaker 2: some biology and And maybe it looks okay, but it's
9:01
Speaker 2: not extraordinary. And we just wind the program down. We're
9:05
Speaker 2: really looking for those resilient programs that we could build
9:07
Speaker 2: a lot of evidence around when we're trying to figure
9:10
Speaker 2: out the translational direction that we needed to go. Things
9:13
Speaker 2: that just kept on sticking. And what I mean by that,
9:15
Speaker 2: we kept on getting good animal work. When we would
9:18
Speaker 2: go into candidate nomination studies, we kept on finding exciting
9:21
Speaker 2: profiles for those molecules. As they're just being very disciplined
9:24
Speaker 2: in terms of we're going to cast a lot of netts.
9:27
Speaker 2: but we're really only going to bear down on the
9:29
Speaker 2: net that just is bountiful. Like we keep on pulling
9:32
Speaker 2: up a feast from the ocean. And that's, that's been
9:34
Speaker 2: the rock story for us. And just making sure that
9:36
Speaker 2: we were lean, we were efficient. We didn't have a
9:39
Speaker 2: lot of bloat on the, on the staff side. We
9:43
Speaker 2: really only hired people that filled the key deficiency in
9:45
Speaker 2: the company. Uh, so things like regulatory, uh, translational pharmacology, uh,
9:51
Speaker 2: all the chemistry that you need in order to work
9:53
Speaker 2: on small molecules. in terms of CMC and the regulatory
9:57
Speaker 2: side of things. And so that's what's been able on
10:02
Speaker 2: the strategy side for us to really move the needle here.
10:05
Speaker 2: On the technology side, what we do is something very
10:08
Speaker 2: different from a lot of machine learning and AI out there.
10:12
Speaker 2: In fact, when we were originally starting the company, Everybody
10:15
Speaker 2: was telling us, oh, you're overfitting, right? And that's the
10:18
Speaker 2: worst thing that you could do in machine learning. It's
10:22
Speaker 2: heresy to use very small data sets and say, no,
10:25
Speaker 2: we're going to have better generalization than these ultra large
10:28
Speaker 2: libraries looking at the universe of chemical space. And now
10:33
Speaker 2: we've got some, we published a paper on this earlier
10:36
Speaker 2: this year, but we've been wrestling with bringing in causality
10:40
Speaker 2: into machine learning. Causality is this discipline of how do
10:44
Speaker 2: we mathematically describe cause and effect, okay? And so let's
10:47
Speaker 2: say I have some data set for a target of interest. Oftentimes,
10:53
Speaker 2: I actually have a bunch of different submechanisms in that set.
10:56
Speaker 2: And from a chemistry and pharmacology perspective, it doesn't make
10:59
Speaker 2: any sense to treat those as if they're mutually informative.
11:02
Speaker 2: But the machine learning community would because it was the target, right?
11:06
Speaker 2: And so you build these very, very large models, and
11:10
Speaker 2: they'd have a good general sense of affinity and potency,
11:14
Speaker 2: but they would lose, every time we tested it on subscaffolds,
11:17
Speaker 2: we would lose the specific details. So it'd be very
11:21
Speaker 2: good at the general, but very bad at the specific.
11:23
Speaker 2: And so what we found is that actually if we
11:26
Speaker 2: focused in, and focus is a concept that we created
11:30
Speaker 2: to describe the balance between fitness and mechanistic resolution or focus,
11:35
Speaker 2: if we focused in on the scaffold, we could become
11:38
Speaker 2: very powerful on that scaffold and basically ignore everything else.
11:43
Speaker 2: And this is very important if you're going to do
11:46
Speaker 2: candidate design, because oftentimes when you've got some cell-based talks
11:50
Speaker 2: or some cell-based off-target pharmacology that you have to model,
11:54
Speaker 2: there are 10, 20, 30, who knows how many mechanisms
11:57
Speaker 2: in that data set and they're hidden and you can't
11:59
Speaker 2: pull them out easily. And so we were able to
12:00
Speaker 2: step in, focus in on that mechanism for that space,
12:04
Speaker 2: build four or five algorithms that can optimize the space
12:07
Speaker 2: And we made 11 compounds to find two clinical candidates.
12:10
Speaker 2: So that's extraordinary, right? That level of efficiency hasn't been
12:15
Speaker 2: obtained by any other group outside of Avicenna. Normally, you
12:18
Speaker 2: have to make hundreds of molecules, thousands of molecules to
12:21
Speaker 2: find a clinical candidate. We made 11 to find two.
12:25
Speaker 2: And we did it again in CSF1R.
12:28
Speaker 1: From the layman's perspective, I want to say this back
12:30
Speaker 1: to you and see if the analogy I'm drawing is right.
12:34
Speaker 2: The.
12:36
Speaker 1: Proverbial wisdom would be with you take the broadest possible
12:40
Speaker 1: data set, you zoom out as much as possible, and
12:43
Speaker 1: you let the AI go to work. That's using the
12:48
Speaker 1: biggest possible data set. And what you're saying is you
12:51
Speaker 1: took a rifle shot more approach and said there's too
12:55
Speaker 1: much noise in all this data. I'm actually going to
12:58
Speaker 1: zoom in. And that's where this is the heresy you're
13:01
Speaker 1: referring to that actually said, by zooming in on what
13:05
Speaker 1: I really want to be intentional about and putting the
13:08
Speaker 1: ML and the algorithm to work, I've actually cut out
13:11
Speaker 1: of a lot of noise. And that's why I'm getting
13:12
Speaker 1: two candidates out of 11 design.
13:16
Speaker 3: Like, is that a fair way to say it back
13:18
Speaker 3: to you?
13:19
Speaker 2: That's exactly right. And being able to inferentially zoom in
13:24
Speaker 2: on the mechanism, that's why we call it inferential mechanics.
13:27
Speaker 2: That's the really exciting thing is it gives us a
13:29
Speaker 2: new mathematical framework to navigate the mechanistic landscape in the
13:33
Speaker 2: natural sciences. And that's really cool from a philosophical perspective.
13:38
Speaker 2: But in terms of drug design, when you build these
13:41
Speaker 2: very general, very nonspecific algorithms, they're actually really weak on
13:45
Speaker 2: a case-by-case-by-case basis. You can think of it as like
13:48
Speaker 2: I've got a pink series that I want to model.
13:51
Speaker 2: And then I've got red, orange, yellow, green, blue, purple
13:54
Speaker 2: in my data set. If I lump them all together,
13:56
Speaker 2: the average color of that is brown, right? Brown doesn't
13:59
Speaker 2: actually exist in my data set. Brown is an artificial
14:02
Speaker 2: color that's the average of all of these. And so
14:04
Speaker 2: I build this brown algorithm, and when I test train,
14:06
Speaker 2: I keep on getting brown because I'm mixing all of
14:08
Speaker 2: these colors. But when I go and make a prediction
14:11
Speaker 2: in the pink space, the red space, the blue space,
14:13
Speaker 2: it doesn't work. when I test just in those regions
14:17
Speaker 2: rather than as an average, then that's the issue.
14:20
Speaker 3: That's so well put.
14:21
Speaker 1: And it is, it's really cool to have you on
14:23
Speaker 1: the show because this is groundbreaking stuff between your paper.
14:26
Speaker 1: We'll get to your op-ed in a second, but you
14:29
Speaker 1: did something different. And in gosh, not five years, but
14:33
Speaker 1: in five months, people are going to be trying to
14:34
Speaker 1: replicate what Avicenna has done and say, maybe we are
14:38
Speaker 1: thinking about this wrong. So you are the right guy
14:39
Speaker 1: for the right moment, because as we speak, as we're recording, um,
14:44
Speaker 1: folks are gathering in Washington to talk about AI and
14:47
Speaker 1: we'll see how much biotech comes up. I don't think
14:49
Speaker 1: it'll come up much, but you penned an incredible piece
14:53
Speaker 1: that is linked to in the comments and the description.
14:56
Speaker 1: So people, please, if you're interested in how AI is
15:00
Speaker 1: reshaping this industry, it is the right piece to read.
15:03
Speaker 1: And in it, you say that AI will change drug
15:07
Speaker 1: design development, but quote, not in the ways widely being presented.
15:12
Speaker 3: Can you spell that out here?
15:14
Speaker 1: What are folks getting wrong with their assumptions of how
15:18
Speaker 1: AI is going to be transformative?
15:20
Speaker 2: I mean, you see it just over and over, even
15:22
Speaker 2: beyond biotech, just the sensationalism is everywhere. And I think
15:27
Speaker 2: that's a lot. That's the point. Number one is we
15:30
Speaker 2: just kind of dial down the sensationalism. A lot of
15:32
Speaker 2: people will actually, a very small number of people stand
15:35
Speaker 2: to make a lot of money off of these sensational claims.
15:38
Speaker 2: And we should be perhaps most skeptical around that group.
15:41
Speaker 2: But In terms of AI and what we're doing in
15:45
Speaker 2: the natural sciences and the clinical sciences, they draw on
15:49
Speaker 2: these knowledge graphs, like these very large frontier models, foundation
15:53
Speaker 2: models that underpin large language models like ChatGPT or CLOD.
16:00
Speaker 2: They go into the literature and you can basically generate
16:04
Speaker 2: these knowledge graphs around literature teachings and abstracts and then
16:09
Speaker 2: full bodies of papers. Sadly, the fundamental issue, or at
16:13
Speaker 2: least one of the fundamental issues, is that a lot
16:15
Speaker 2: of the data that underpin the literature is irreproducible. And
16:20
Speaker 2: we've known about this for about 20 years in the
16:22
Speaker 2: natural sciences. And it's been a big issue, especially in
16:25
Speaker 2: more applied sciences like psychology, where straight up you just
16:29
Speaker 2: can't reproduce the findings of the paper. And so... That's
16:33
Speaker 2: a major issue if you treat that as if it's
16:36
Speaker 2: true knowledge and you start learning trends from a knowledge
16:40
Speaker 2: graph across all of these irreproducible results. Well, you're going
16:44
Speaker 2: to learn things that aren't actually real, right? And in
16:49
Speaker 2: terms of biotech, a great example of this is the
16:51
Speaker 2: beta amyloid story. It turns out, I think this was
16:54
Speaker 2: now three or four years ago, a lot of the
16:57
Speaker 2: original beta amyloid work linking beta amyloid pathology to Alzheimer's
17:03
Speaker 2: was just completely fabricated. And so this has been a
17:06
Speaker 2: major problem for Alzheimer's research because one of the core
17:10
Speaker 2: bedrocks of the discipline now is no better than sand.
17:15
Speaker 2: We can't build on this foundation. And that's just one
17:18
Speaker 2: example of many, many, many examples where You go in
17:21
Speaker 2: and you try and replicate something and it's just, it's
17:24
Speaker 2: not even wrong, right? It's nowhere close to even being
17:27
Speaker 2: sort of kind of incorrect. It's just completely irreproducible in noise.
17:32
Speaker 2: That's a lot of what is being touted right now
17:36
Speaker 2: in AI is the ability to sift through the literature,
17:38
Speaker 2: be able to sift through internal data sets and find
17:40
Speaker 2: novel biological connections. And we just aren't at the level
17:44
Speaker 2: of resiliency and redundancy in our literature in order for
17:48
Speaker 2: that to truly be a revolution, to be transformative. So
17:52
Speaker 2: that's point number one. Point number two is that drug design,
17:59
Speaker 2: especially if you're dealing with new biology and new chemistry,
18:03
Speaker 2: you don't know what's going to work until you're on
18:05
Speaker 2: the other side of a new drug application, right? So
18:10
Speaker 2: that's the problem is there's a 10-year lag between, wow,
18:13
Speaker 2: this looks really cool and maybe we should invest some
18:15
Speaker 2: time and and resources in this to actually becoming a drug.
18:19
Speaker 2: And even then you go to phase four, right? Drugs
18:21
Speaker 2: come off market because they aren't actually working, right? So
18:25
Speaker 2: that's the other problem is the cycle for knowledge in
18:29
Speaker 2: translational pharmacology, clinical pharmacology is only order of a decade.
18:34
Speaker 2: And all of these, I've had so many ideas in
18:39
Speaker 2: Avicenna with my colleagues that, man, this looks really cool.
18:42
Speaker 2: And then we go and actually dive into it and
18:45
Speaker 2: look at the translational pharmacology and it turns out it's
18:47
Speaker 2: just completely wrong. Good ideas are cheap. Durable ideas are
18:52
Speaker 2: the hard part. And AI can't help you there.
18:54
Speaker 1: That– well, I think I know what tagline the production
18:57
Speaker 1: team is going to pull. Those good ideas– I want
19:01
Speaker 1: to stay on this for one second because I want
19:03
Speaker 1: to make sure listeners are picking this up. I think
19:08
Speaker 1: there is a belief that we can set AI loose
19:13
Speaker 1: on the last century of academic literature, basic research, NIH,
19:18
Speaker 1: all the way up to true drug discovery, and it's
19:22
Speaker 1: going to just find things that we missed. I think
19:25
Speaker 1: your piece and what you just said is calling out
19:28
Speaker 1: something that we don't like to talk about, that there
19:31
Speaker 1: are some real problems in the data sets going back
19:34
Speaker 1: the last few decades, and they're not reproducible, and if
19:38
Speaker 1: you're training your AI on things that aren't borne out
19:42
Speaker 1: in actual fact.
19:43
Speaker 3: And so is part of what we have.
19:46
Speaker 1: To do to really maximize this, and you do some
19:49
Speaker 1: math that I want you to get into. I think
19:52
Speaker 1: delivering 100 years of new drugs in the next 10,
19:57
Speaker 1: I think is a big claim. And you kind of
20:01
Speaker 1: take it to task in your piece. I want you
20:03
Speaker 1: to unpack that. but is also part of this. We've
20:07
Speaker 1: got to clean up a lot of the last 20,
20:10
Speaker 1: 30 years of science and these assumptions. And we've got
20:13
Speaker 1: to get serious. And this goes back to your rifle shot.
20:17
Speaker 1: We've got to be serious about what we're putting in,
20:19
Speaker 1: not building on a foundation of sand. We've got to
20:21
Speaker 1: build on a foundation of stone. And right now, how
20:26
Speaker 1: do we get that confidence level up that what we
20:29
Speaker 1: took as fact for the last 20 years is actually fact?
20:33
Speaker 2: Well, and I want to extend it just a little bit.
20:35
Speaker 2: Like even good papers, good science can go on and
20:40
Speaker 2: people find out actually there were hidden things all along
20:43
Speaker 2: that were driving that conclusion. And that's the issue is
20:47
Speaker 2: like there are really, really extraordinary stories. Like I'll give
20:51
Speaker 2: you an example rather than speaking in platitudes. But so
20:54
Speaker 2: there's this chemistry in this. I apologize if you're not
20:57
Speaker 2: a chemist, going to nerd out a little bit, but
20:59
Speaker 2: there's a reaction called the Nozaki-Hiyama-Kishi reaction. And it came
21:04
Speaker 2: around in the 70s. And it was this really cool
21:08
Speaker 2: way to couple vinyl halides with carbonyls in one pot. Normally,
21:14
Speaker 2: this is really hard. You have to use some very
21:16
Speaker 2: dangerous chemistry in order to do this. But people found
21:19
Speaker 2: that there was this chromium-mediated reaction that could facilitate this coupling.
21:25
Speaker 2: And lo and behold, people tried to replicate it. And
21:28
Speaker 2: they couldn't. They were using the same chromium reagent, same substrates,
21:32
Speaker 2: couldn't get the reaction to work. And it turns out
21:35
Speaker 2: the original chromium samples that were being used for this
21:38
Speaker 2: were contaminated by nickel. It was actually a chromium-mediated nickel-catalyzed reaction.
21:44
Speaker 2: And so the moment this other group put nickel into
21:47
Speaker 2: their reaction, the Nozaki-Hiyama-Kishi reaction would work. No nefariousness, no sloppiness,
21:53
Speaker 2: just a trace impurity in a chromium sample was the
21:57
Speaker 2: serendipity behind that discovery. And that story is very common, right?
22:01
Speaker 2: The thing that you become comfortable with, comfortable is perhaps
22:05
Speaker 2: not the right word, but the thing that you become
22:08
Speaker 2: familiar with in the natural sciences is how little we
22:12
Speaker 2: know about nature and how complex the universe is. And
22:17
Speaker 2: it's a rare thing when you truly can find some
22:20
Speaker 2: new knowledge that's durable. And even when you do everything right,
22:24
Speaker 2: and you're very careful, considerate, and deliberate there's still a
22:27
Speaker 2: massive role of serendipity in discovering. There always will be
22:32
Speaker 2: because the universe is just that complex. And so that's
22:34
Speaker 2: the thing that the computer scientists don't fully understand is
22:39
Speaker 2: just the stupefying amount of complexity in chemistry, biology, and physics.
22:44
Speaker 2: And that's something that chemists, biologists, and physicists understand very,
22:47
Speaker 2: very clearly. So unless you're working in well-established domains where
22:52
Speaker 2: there's a lot of reproducibility, and you know all of
22:55
Speaker 2: the unknowns more or less, I really can't help you
22:58
Speaker 2: because the universe is just that unknown, right? I mean,
23:02
Speaker 2: you know, I'm gesticulating with my hand. Think about how
23:04
Speaker 2: many Nobel prizes are going on in my hand and
23:06
Speaker 2: I just use it routinely to make gestures. All of
23:09
Speaker 2: those aspects of biology are still important for the day-to-day
23:12
Speaker 2: functioning of my hand and its physiology. We don't know
23:16
Speaker 2: about those aspects. They're just in the background, right? And
23:20
Speaker 2: that's the hard part is how do I design artificially
23:24
Speaker 2: intelligent systems, pattern recognition systems, machine learning agents. How do
23:28
Speaker 2: I design those things to be able to find design
23:33
Speaker 2: goals or new knowledge in that background of uncertainty and
23:37
Speaker 2: irreproducibility and serendipity? That's the hard part.
23:41
Speaker 1: This is definitely going to be an episode of humility
23:43
Speaker 1: because I think you're laying it out.
23:45
Speaker 3: There's so much we don't know.
23:47
Speaker 1: And if something AI does well is it manipulates what
23:51
Speaker 1: we know for certain allows us to apply it in
23:54
Speaker 1: new ways. But if we're not certain about what we know,
23:57
Speaker 1: there are some limitations, which comes up with, what do
23:59
Speaker 1: you make of the claim that we're going to do
24:01
Speaker 1: 100 years of biotech in the next 10 years?
24:05
Speaker 2: Yeah, it's sensationalist, right? That's the word of the day
24:13
Speaker 2: for a lot of these these AI companies is they're
24:16
Speaker 2: making these sensational claims and it's catching a lot of
24:18
Speaker 2: attention and it sounds cool and people want it to
24:21
Speaker 2: be cool. So they're raising huge piles of money, but
24:24
Speaker 2: it's just not going to be durable. And that's the,
24:27
Speaker 2: the issue is we would have to discover, depending upon
24:31
Speaker 2: your estimate, how you run the arithmetic between 450, 550
24:34
Speaker 2: new drugs every year in order for that to be true.
24:36
Speaker 2: And what, That's just an unfathomable amount of drug development work.
24:49
Speaker 2: If we run the capital requirements of that, you and
24:51
Speaker 2: I had a conversation via email with Grace, and I
24:55
Speaker 2: think John Guy asked the question, so you think you're
25:00
Speaker 2: going to saturate the regulatory environment? No, we're going to
25:04
Speaker 2: saturate the capital markets before that, right? 500-some-odd drugs is
25:09
Speaker 2: a lot of money, right? There's no way we have
25:11
Speaker 2: enough money around to develop those, let alone the scientific
25:14
Speaker 2: know-how to do that at that scale. It's just an
25:17
Speaker 2: absurd number.
25:19
Speaker 3: It's such a good call.
25:21
Speaker 1: I mean, you see the numbers around CapEx on data
25:24
Speaker 1: centers and AI, and you think the pot is limitless.
25:27
Speaker 1: And if you have a good idea, the capital is
25:30
Speaker 1: going to be there. And you make the point, we're
25:32
Speaker 1: not going to 10x VC overnight. We're not going to
25:34
Speaker 1: go from $ 40 billion to $ 400 billion overnight. That's ludicrous,
25:40
Speaker 1: let alone– I love– our policy director, John Guy, asks
25:43
Speaker 1: the best questions. Is the regulatory ecosystem prepared for 10x
25:48
Speaker 1: the number of submissions? And I think you make the
25:51
Speaker 1: good point that we'll be capitals constrained before we're regulatorily constrained.
25:56
Speaker 1: So that's– so there are nuances. It's not– AI is
26:00
Speaker 1: not going to be the easy button. But let's close
26:03
Speaker 1: out AI on a positive note. Where do you see
26:06
Speaker 1: it helping and where do you see it being transformative?
26:09
Speaker 2: Yes. So two areas. One that is within Avicenna's remit
26:13
Speaker 2: is clinical candidate design. If you have a very good
26:17
Speaker 2: idea of your biology, your translational pharmacology, and you just
26:20
Speaker 2: need to put all the pieces together on your medicinal chemistry,
26:23
Speaker 2: AI is very powerful there, right? We've demonstrated this. It's
26:27
Speaker 2: transformative tools. because it can solve that multi-objective optimization problem
26:31
Speaker 2: for you very efficiently, if and only if you know
26:35
Speaker 2: what your target product profile is, right? And you know
26:39
Speaker 2: more or less how you can massage the chemistry in
26:41
Speaker 2: order to match up with that. The other area that
26:44
Speaker 2: I think is really exciting is being able to sift
26:47
Speaker 2: through electronic health records and find either repurposing, repositioning opportunities
26:54
Speaker 2: where we can redesign new candidates, be they small molecule
26:58
Speaker 2: or biologic. And then also finding associations between diseases and
27:05
Speaker 2: serological markers and maybe gleaning some new pathology and some
27:09
Speaker 2: new disease biology there. There's a lot of work there
27:13
Speaker 2: because you want to talk about noisy data sets for
27:16
Speaker 2: training and learning. Electronic health records are very noisy and
27:19
Speaker 2: they're often very fragmented and it's going to be a
27:22
Speaker 2: big lift to get them ready for machine learning. But
27:26
Speaker 2: already you're seeing groups learning, hey, I've got some clinicians
27:31
Speaker 2: that have been reusing this drug off-label for this indication
27:34
Speaker 2: that we thought was completely unrelated, and it's been working.
27:37
Speaker 2: How do we design a drug to go after that
27:38
Speaker 2: indication and that target? Maybe there's some new pharmacology there
27:43
Speaker 2: for us to work on. So that, I think, is
27:46
Speaker 2: when you can constrain the context and you can ask
27:51
Speaker 2: very deliberate questions about pharmacology pathology, And clinical trial design.
27:59
Speaker 2: I think AI is going to be very powerful. But
28:01
Speaker 2: you have to be very deliberate about asking your question.
28:05
Speaker 2: And you have to make sure that the context underneath
28:08
Speaker 2: your feet isn't sifting unwittingly. That's where AI becomes very bad.
28:12
Speaker 2: If the contextual rules change and you don't know they've changed,
28:16
Speaker 2: then AI is no better than guessing.
28:20
Speaker 1: This has been the bee in my bonnet about this
28:23
Speaker 1: because I think of this also in a globally competitive standpoint.
28:27
Speaker 1: I am concerned– so you tell me whether this is
28:30
Speaker 1: a worry you share– that our fragmented data sets held
28:35
Speaker 1: by hospitals, held by insurers, the EHR revolution didn't really
28:41
Speaker 1: materialize I think the way we thought it might. Our data,
28:46
Speaker 1: patient data– failed clinical trials, they all sit at different
28:51
Speaker 1: entities that'll have different ways most likely to monetize that data.
28:58
Speaker 1: I had an interesting conversation with some of the leadership
29:00
Speaker 1: in the UK on their drug development side. They have
29:04
Speaker 1: centralized data. They can just tap into the NHS. You know,
29:08
Speaker 1: people look to where does Australia get an advantage? Where
29:11
Speaker 1: does Germany get an advantage? Where does China get an advantage?
29:14
Speaker 1: That's certainly the number one topic. Are you concerned that
29:20
Speaker 1: the entities that hold our data here, especially on the
29:24
Speaker 1: insurer side and the hospital side, will make it too
29:27
Speaker 1: difficult or too slow to aggregate that fragmented data in
29:30
Speaker 1: a way that we're competitive vis-a-vis the NHS?
29:34
Speaker 2: Oh, yeah. So I went to Oxford for medical school
29:39
Speaker 2: and earned my license to practice. in the UK. And
29:42
Speaker 2: that training under a single-payer healthcare system opened my eyes
29:47
Speaker 2: to a lot of things that we are doing not
29:49
Speaker 2: quite right in the US. But I was just up
29:52
Speaker 2: in Canada at the Gardner Lecture for AI and Healthcare,
29:56
Speaker 2: and And I gave a talk on influential mechanics and
30:00
Speaker 2: what Avacyn is doing. But the thing that really caught
30:02
Speaker 2: my ear is how the province of Alberta and Canada
30:05
Speaker 2: in general are thinking about electronic health records. It's centralized.
30:09
Speaker 2: They're making it accessible where there's one regulatory process for
30:14
Speaker 2: accessing that. They're very careful about patient safety and patient privacy.
30:18
Speaker 2: But beyond that, there's a centralized portal to access all
30:22
Speaker 2: of this information. And it's interesting to me that the
30:24
Speaker 2: thought that I had was, Wouldn't it be ironic if
30:29
Speaker 2: data monetization was the thing that pushed the U.S. to
30:31
Speaker 2: a national healthcare system? Like, we've got all this private data.
30:35
Speaker 2: We've got to make some money off of it, so
30:36
Speaker 2: let's have a single-payer system. That would be ironic. But, yeah,
30:42
Speaker 2: I think you're exactly right. The frontier for discovery in
30:47
Speaker 2: terms of new drive opportunities, new targets... new ways of
30:52
Speaker 2: doing patient recruitment or adaptive clinical trial design, especially with
30:56
Speaker 2: AI where you're feeding the information, serological markers, and you're
31:00
Speaker 2: literally adapting the trial as you run it to the
31:02
Speaker 2: response of patients. That's going to require a very savvy
31:07
Speaker 2: and sophisticated EHR processing system. And those national healthcare systems
31:13
Speaker 2: that have unified single point of access EHRs, are going
31:18
Speaker 2: to beat us because our system is so fragmented. Like
31:20
Speaker 2: if I want to go down the street, let's say
31:22
Speaker 2: I want to go to Medical University South Carolina, I
31:25
Speaker 2: have to go and interact with a very different framework
31:28
Speaker 2: than I would if I want to go down the
31:31
Speaker 2: street in Durham to Duke or to UNC Chapel Hill.
31:36
Speaker 2: And that's just three universities within a relatively small geographic
31:39
Speaker 2: region of the United States. It's a nightmare to access
31:43
Speaker 2: large-scale patient repositories compared to what Britain has to offer
31:47
Speaker 2: or what Canada is building. Mark Carney was up on
31:49
Speaker 2: the stage when they were rolling out the Canada initiative
31:53
Speaker 2: for AI. First and foremost, it was healthcare. And it
31:56
Speaker 2: was around building AI for all in healthcare. And so
32:00
Speaker 2: that's a major initiative in Canada right now is to
32:04
Speaker 2: really take advantage of the national advantage that they have
32:07
Speaker 2: in terms of EHR and other similar things and lead
32:11
Speaker 2: the way for AI in healthcare, be it AI scribing,
32:15
Speaker 2: AI in adaptive clinical trial design, et cetera. And we
32:19
Speaker 2: just don't have the infrastructure on the informatics side in
32:22
Speaker 2: the US to compete with that right now.
32:26
Speaker 1: When you have a genus on the show, you throw
32:28
Speaker 1: out as many questions as you can.
32:30
Speaker 3: Okay.
32:31
Speaker 1: Is the data abroad good enough for someone like yourself
32:38
Speaker 1: to go ping it there? Or do we need to
32:41
Speaker 1: build this here in the U.S.
32:43
Speaker 2: Yeah, so the answer is both, and it really does depend.
32:49
Speaker 2: The interesting thing about, like, if you've got very rare
32:53
Speaker 2: diseases with a very uniform genetic background, the data abroad
32:57
Speaker 2: are probably going to be very useful for us. Like,
33:01
Speaker 2: von Lippenau mutations are going to behave the same largely
33:05
Speaker 2: in Britain versus in the US. But one of the
33:10
Speaker 2: interesting things that we've seen, and this kind of goes
33:13
Speaker 2: to Japanese regulatory policy, is sometimes drugs behave in populations
33:17
Speaker 2: very differently than other populations at a national scale. In fact,
33:20
Speaker 2: Japan insists that a drug be tested on Japanese people
33:24
Speaker 2: in order to get regulatory approval in Japan. And that's
33:26
Speaker 2: because the genetic background shifts, the cultural background shifts, the food,
33:30
Speaker 2: the exercise habits shift. So there's a lot of extra
33:34
Speaker 2: stuff that goes into the background of a drug working,
33:38
Speaker 2: that sometimes if you shift nation by nation by nation,
33:41
Speaker 2: you see different response rates for drugs that work via
33:46
Speaker 2: multiple insults or multiple genetic backgrounds that give rise to
33:50
Speaker 2: that disease. So it really depends. For single mutation rare diseases,
33:58
Speaker 2: I think that a national agnostic data set will be
34:02
Speaker 2: very useful. But for things like metabolic syndrome, general colon
34:08
Speaker 2: cancer risk, inflammatory immunology indications that have a lot of
34:13
Speaker 2: different reasons for the disease appearing. Those are still going
34:18
Speaker 2: to require a national level system in order to be
34:21
Speaker 2: able to address those diseases on the terms of the
34:23
Speaker 2: United States.
34:24
Speaker 1: To say that back to you another way, the more
34:26
Speaker 1: environmental factors might play a role in a disease, we're
34:30
Speaker 1: going to want to have a U.S.
34:31
Speaker 3: Data set. And I think that ties.
34:33
Speaker 1: Very much into the Maha, what we eat, our food
34:36
Speaker 1: ecosystem is different here. And so things that are modified
34:40
Speaker 1: by what you surround yourself with. That's where we're going
34:43
Speaker 1: to wish we had a U.S. population data set.
34:45
Speaker 2: Exactly so.
34:46
Speaker 3: Super, super helpful.
34:47
Speaker 1: You mentioned Japan, and it's actually very timely, one, because
34:51
Speaker 1: Making Medicine will be on the road in Japan in
34:54
Speaker 1: a week's time. But while we're there, we're going to
34:57
Speaker 1: be talking to a lot of their national leaders who
35:00
Speaker 1: continue to ask why they have drug lag and drug
35:03
Speaker 1: loss over there. And you point to their regulatory system.
35:08
Speaker 1: I think we're optimistic that they're also going to start
35:10
Speaker 1: valuing medicine in a more robust manner when it demonstrates
35:16
Speaker 1: that it does have that value for the Japanese population.
35:20
Speaker 1: When you balance– putting on your investor hat, when you
35:23
Speaker 1: balance something like that, is it both the pricing and
35:28
Speaker 1: reimbursement as well as the regulatory system or do they
35:33
Speaker 1: need to pull both levers? Make it easier to get
35:35
Speaker 1: a drug approved in the market. but you also have
35:38
Speaker 1: to fairly value and reimburse it.
35:40
Speaker 2: I think short answer, yes. And there's a third thing.
35:43
Speaker 2: The venture ecosystem in Japan is very small. So like
35:47
Speaker 2: a big Series B is $ 6 million. So it makes
35:52
Speaker 2: it very difficult to actually bring drugs to market that
35:56
Speaker 2: are fully incubated in Japan in the biotech sector. Don't
35:59
Speaker 2: get me wrong, Takeda and Otsuka and others, they're very
36:03
Speaker 2: good at large-scale traditional pharma drug design development. But the
36:09
Speaker 2: biotech ecosystem in Japan is still quite small. Regulatory path
36:13
Speaker 2: to approval is a major issue and then absolutely reimbursement.
36:17
Speaker 2: People aren't going to invest in your drug program if
36:19
Speaker 2: they think, well, payers aren't going to pay. So that's
36:23
Speaker 2: a major issue for ultra-rare indications is how do we
36:26
Speaker 2: make sure that they're investable because there are areas of
36:29
Speaker 2: huge need that are still poorly served. But I think
36:33
Speaker 2: it's a constellation of all three, that the biotech ecosystem
36:36
Speaker 2: is still fairly small, especially on the investment side. It's
36:38
Speaker 2: difficult to get regulatory approval in Japan. And you've got
36:42
Speaker 2: to make sure that there is appropriate remuneration for that
36:45
Speaker 2: investment risk.
36:46
Speaker 1: I'm going to keep throwing them at you. We're going
36:48
Speaker 1: to take full advantage of the time you've given us.
36:50
Speaker 1: Let's move a little south in the Pacific down to China.
36:53
Speaker 1: Hot topic about how Chinese rise in the biotech space
37:00
Speaker 1: is transformative. And we certainly have pockets of protectionism within
37:05
Speaker 1: our own industry. We've got what's treasury going to do
37:07
Speaker 1: with coins.
37:08
Speaker 3: We've got, is.
37:09
Speaker 1: Congress going to do more? How do you, someone who
37:13
Speaker 1: both sits on the founder side, now the investment side,
37:17
Speaker 1: you're one of the big minds in AI. How do
37:20
Speaker 1: you think about China in the global drug development landscape?
37:24
Speaker 2: Yeah, we've been training Chinese scientists in the United States
37:27
Speaker 2: and the West in general for 50 years. So it's
37:30
Speaker 2: not surprising to me that talented people are now going
37:34
Speaker 2: to go back to their country and start competing with
37:38
Speaker 2: us on our terms. The Chinese are extraordinarily talented at
37:44
Speaker 2: small molecule chemistry, organic synthesis, organometallics. So it's only natural
37:49
Speaker 2: that I think they'll lean into those industries, especially when
37:52
Speaker 2: they're backed by the Chinese government. One of the things
37:56
Speaker 2: that I still think we do better than they do
38:00
Speaker 2: is completely new targets, optimization, and then translational pharmacology. So
38:07
Speaker 2: we still have a huge biological innovation competitive advantage here.
38:11
Speaker 2: And there's a lot of risks there, as I alluded
38:13
Speaker 2: to earlier. But when you do find new targets that
38:16
Speaker 2: can translate... that's really where we can keep a significant
38:21
Speaker 2: competitive edge. And then also, I think going after rare
38:26
Speaker 2: diseases is something that the U.S. will retain a competitive
38:30
Speaker 2: edge in because those clinical trials are often very atypical.
38:34
Speaker 2: They require a lot of expertise to plan, especially for
38:37
Speaker 2: rare CNS indications. There's often two acronyms, PK and PD,
38:45
Speaker 2: pharmacokinetics and pharmacodynamics. The PKPD linkage in CNS is very
38:51
Speaker 2: hard to address because you're modulating targets deep in the brain.
38:55
Speaker 2: Often they're intracellular. And so you have to be very,
38:58
Speaker 2: very good about addressing that PKPD response and overlap as
39:04
Speaker 2: part of your translational pharmacology package. And that still requires
39:07
Speaker 2: a lot of expertise that industry in the US still
39:10
Speaker 2: has compared to China.
39:12
Speaker 1: Something when I sit with leaders in the rare disease
39:14
Speaker 1: community and hear about their priorities, The IITs in China,
39:22
Speaker 1: which China is through their 818 process beginning to corral,
39:27
Speaker 1: I would say, some of these more out there IITs. Obviously,
39:31
Speaker 1: we kind of have started to see some tragic stories
39:34
Speaker 1: coming out of China where clinical trial design may have
39:38
Speaker 1: played a role from safety. But then I also hear
39:41
Speaker 1: from rare disease advocates that they could only move an
39:45
Speaker 1: asset forward because of the low cost of these investor
39:50
Speaker 1: investigator initiated trials. How do you think about IITs in China? Um,
39:56
Speaker 1: because it seems very nuanced. There are some benefits of
39:59
Speaker 1: moving something forward very cheaply. Um, but there are also
40:02
Speaker 1: probably some drawbacks that these trials wouldn't be allowed to
40:06
Speaker 1: be stood up or they wouldn't be stood up in
40:08
Speaker 1: similar ways in the U S.
40:10
Speaker 2: Yeah, I completely agree. I mean, I, I think the
40:14
Speaker 2: fact that so many, even investors that are making a
40:19
Speaker 2: big China play, they still want to repeat these pivotal
40:22
Speaker 2: trials in the U.S. under U.S. supervision tells you a
40:25
Speaker 2: lot about how much confidence even Western investors that are
40:32
Speaker 2: in the China hypothesis, how much confidence they have in
40:35
Speaker 2: the data that come out of those trials. You really
40:37
Speaker 2: want to make sure that under very rigorous, highly controlled terms,
40:41
Speaker 2: The pharmacology that you see is the pharmacology that was
40:44
Speaker 2: seen in China. And I think there's a risk there
40:49
Speaker 2: that you can go too fast, too quickly with addressing
40:52
Speaker 2: all of your confounders, going back to causality. That was
40:56
Speaker 2: one of the early areas of bringing causality into the
41:01
Speaker 2: sciences was actually clinical trials and clinical data and addressing
41:04
Speaker 2: these confounders. If you go too quickly, oftentimes you can
41:07
Speaker 2: have confounders that come along for the ride or, as
41:10
Speaker 2: you were talking about, there are big safety issues where
41:12
Speaker 2: you didn't appropriately recruit. You allowed people to come in
41:16
Speaker 2: who were actually fairly high risk to your target modification
41:19
Speaker 2: and they got end organ damage or something worse, death.
41:23
Speaker 2: And that's something that I think the regulatory side in
41:27
Speaker 2: the U.S. is very good at, especially in phase two
41:30
Speaker 2: and phase three, is patient safety. That is their mantra.
41:33
Speaker 2: Your efficacy is your nickel, but FDA comes into the
41:36
Speaker 2: discussion around patient safety. And I think the areas where
41:41
Speaker 2: we can learn, especially if we look at what, you know,
41:44
Speaker 2: Canada also has a fairly slow clinical development trajectory from
41:48
Speaker 2: a regulatory perspective. But if you look at other Western nations,
41:51
Speaker 2: Australia is amazing from a phase one perspective, just because
41:56
Speaker 2: the filing package that you have to file in order
41:59
Speaker 2: to go into phase one is so much less complex
42:03
Speaker 2: than the IND that you have to submit to the FDA.
42:06
Speaker 2: You don't have to have an IND to go into Australia.
42:08
Speaker 2: You don't have to have the final reports. You do
42:11
Speaker 2: have to have your GLP talk stated, and it is
42:13
Speaker 2: reviewed by a committee. But oftentimes, getting those final reports,
42:15
Speaker 2: getting everything certified, going through the INB process, that adds
42:19
Speaker 2: six months of time. And so you can get into
42:22
Speaker 2: Australia in phase one trials very quickly. I think kind
42:28
Speaker 2: of the China-Australia question on how do we accelerate development, Initially,
42:34
Speaker 2: I remember when people were first talking about Australian drug development,
42:38
Speaker 2: people were talking about the tax credit and tax benefits. Now,
42:41
Speaker 2: I think people have realized, it's okay, it's nice, I'll
42:45
Speaker 2: take it on some of my programs potentially. But the
42:48
Speaker 2: real savings is the fact that I can go into
42:51
Speaker 2: my phase one, I can address some early pharmacology questions,
42:56
Speaker 2: maybe some pivotal questions around target engagement, even healthy volunteers.
43:01
Speaker 2: And I can do that three to six months faster
43:03
Speaker 2: than I can in the US. And that's a big deal,
43:06
Speaker 2: especially if you're a small biotech where your burn rate
43:08
Speaker 2: is very sensitive. You think about six months is the
43:12
Speaker 2: difference between failure and the company thriving, right? That's huge.
43:16
Speaker 2: And I think we can learn from that. So there
43:19
Speaker 2: are definitely things that we can learn from the Chinese
43:22
Speaker 2: approach going faster. We have to go faster in a
43:24
Speaker 2: very careful way, making sure that we don't sacrifice patient safety.
43:28
Speaker 2: But We can look at other approaches to see how
43:31
Speaker 2: we can preserve safety and still go fastly into the clinic.
43:35
Speaker 2: could be my full complicated answer.
43:38
Speaker 3: It was a good one.
43:39
Speaker 1: I'm going to throw a follow-up and then one more question,
43:41
Speaker 1: then I promise we'll let you go. But this has
43:43
Speaker 1: been such a terrific conversation. On the concept of patient safety,
43:48
Speaker 1: I was struck by some data that came out from MassBio.
43:52
Speaker 1: They surveyed, I don't know, a lot of biotech leaders.
43:56
Speaker 1: And something that came up that I didn't hear often
43:59
Speaker 1: before is that our threshold for adverse reactions is in
44:03
Speaker 1: the U.S. might be a little too high or low,
44:07
Speaker 1: or depending on how you're looking at it, that that
44:11
Speaker 1: threshold is maybe being triggered too early and that there
44:14
Speaker 1: probably needs to be a little more tolerance for adverse
44:18
Speaker 1: reactions or really how we're reporting adverse reactions. I don't
44:23
Speaker 1: know if you've seen that data set, but it just
44:27
Speaker 1: reminded me of it because you were talking about patient
44:30
Speaker 1: safety as such a north star. Is that a trade-off
44:33
Speaker 1: we need to consider or can we be competitive and
44:37
Speaker 1: still keep that extraordinarily high bar?
44:40
Speaker 2: It really depends on the indication. If we're talking about
44:46
Speaker 2: transient renal impairment in the context of a third-line AML
44:52
Speaker 2: drug where it's curative, I think we can be very
44:55
Speaker 2: comfortable about that because often AML patients tend to be younger.
44:59
Speaker 2: You're going to die, usually, if you can't treat the cancer.
45:05
Speaker 2: And that transient renal impairment, if you go off the
45:07
Speaker 2: drug after you've gone into remission or cure and your
45:11
Speaker 2: kidneys bounce back, it's probably okay. Now, that being said,
45:15
Speaker 2: a transient renal impairment in Parkinson's disease, I'm less inclined
45:21
Speaker 2: to tolerate because it's an older population already. You tend
45:24
Speaker 2: to have some kind of renal impediment just because of
45:27
Speaker 2: the age. So I think it's really case by case
45:30
Speaker 2: in terms of indication, patient population. And I think people
45:34
Speaker 2: are getting smarter at that. Regulators definitely are starting to think, okay,
45:38
Speaker 2: these blanket statements that we make for all patients, all conditions,
45:41
Speaker 2: we need to start updating our guidance and be really
45:44
Speaker 2: thoughtful on what we're demanding from a safety perspective. That
45:49
Speaker 2: I think is happening out there. So that's really what
45:53
Speaker 2: I would say is it really depends on your indication,
45:56
Speaker 2: the kind of therapy that you're offering, and the background
46:00
Speaker 2: patient population in terms of what constitutes truly a red
46:03
Speaker 2: flag or maybe a proceed with caution.
46:06
Speaker 1: I love that point that we probably need to get
46:09
Speaker 1: away from blanket statements. The science and the precision has
46:13
Speaker 1: gotten us to a place where some things make sense
46:15
Speaker 1: over here, may not make sense over here.
46:17
Speaker 3: Last question, you have been an advocate.
46:20
Speaker 1: I've gotten to see you come up to DC to
46:22
Speaker 1: meet with policymakers. And one of the ones you've talked
46:26
Speaker 1: about is the continued small molecule penalty, this pill penalty
46:30
Speaker 1: that is a part of the IRA. This is nine
46:35
Speaker 1: years versus 13, pick whatever you want to call it.
46:38
Speaker 1: Still an issue, still something you see out there, still
46:41
Speaker 1: a problem. And does Congress need to get the EPIC
46:44
Speaker 1: Act done?
46:45
Speaker 2: Yeah, absolutely. You know, we at ALS, we in Epson
46:51
Speaker 2: are working on ALS. You've got a couple of things
46:54
Speaker 2: that present some headwinds occasionally that you're working in something
46:58
Speaker 2: like amyotrophic lateral sclerosis. It's a hard disease, scares a
47:01
Speaker 2: lot of people. And then there's the concern that payers
47:05
Speaker 2: are going to depress your return at the end of
47:09
Speaker 2: nine years because you're a small molecule. And so I
47:13
Speaker 2: think EPIC Act needs to get done. There's still a
47:16
Speaker 2: lot of excitement around it. The last time I was
47:19
Speaker 2: up on the hill was in May and people were
47:21
Speaker 2: talking about how we need to move this forward. We
47:24
Speaker 2: need to remove the the pill penalty. Uh, and I
47:27
Speaker 2: think it's time that we, we get that sealed and, and, and, and,
47:31
Speaker 2: In the rearview mirror, because there are other more pressing
47:32
Speaker 2: things that we need to talk about.
47:35
Speaker 1: I think you're right. We covered a lot of them here.
47:37
Speaker 1: This is going to be one of our standout episodes.
47:40
Speaker 1: I mean, talk about a masterclass on AI in particular.
47:43
Speaker 1: Congratulations to Avicenna. Congratulations on adding an investor hat. We'll
47:50
Speaker 1: have to have you back on soon. Tom, thank you
47:52
Speaker 1: so much for.
47:52
Speaker 3: Being with us.
47:53
Speaker 2: Thank you, John. It's always a pleasure. It's good to
47:55
Speaker 2: be back.
47:56
Speaker 1: That wraps up this episode of the Making Medicine Podcast.
47:59
Speaker 1: A huge thank you to Tom for joining us again.
48:02
Speaker 1: And congratulations to him and his colleagues at Avicenna for
48:05
Speaker 1: the incredible work they're doing in ALS. Be sure to
48:07
Speaker 1: subscribe on YouTube. Follow us on other channels. Thanks for listening.
48:11
Speaker 1: And as always, keep innovating.