February 27, 2025 • 58 mins
“As US biotech companies, how do we continue to differentiate and create novel breakthrough assets? I think it’s leveraging AI,” said Sean McClain, founder and CEO of Absci. McClain and Christian Stegmann, senior vice president of drug creation, sit down with Bloomberg Intelligence analyst Andrew Galler in this episode of the Vanguards of Health Care podcast to discuss Absci’s generative AI platform and developing best-in-class biologics.
See omnystudio.com/listener for privacy information.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:18):
Hi everyone, and welcome to the Vanguards of Healthcare podcast.
I'm Andrew Galler, one of the senior biotech analysts with
Bloomberg Intelligence. Bloomberg Intelligence is the corporate research arm of
Bloomberg that encompasses five hundred analysts worldwide. I'm joined today
by Sean McClain, founder and CEO of ABSI, and Christian Stegman,
Senior Vice president of Drug Creation Shan. Christian, thanks so
(00:40):
much for joining us today.
Speaker 2 (00:41):
Yeah, thanks for having us here. Andrew, thanks for having us.
Speaker 1 (00:44):
Maybe to kick it off, I'm saying it's a bit
relevant in the news. A few weeks ago, an article
from stat News came out claiming you and some of
your peers were mischaracterizing your work as Denovo antibody design
through jen Ai.
Speaker 2 (00:55):
How would you respond?
Speaker 3 (00:57):
Yeah, so, I guess I'll start off with the the
thing that I agree most with with the staff reporter. Uh,
two years ago, definitely was using too much hair jail,
So hopefully today the hair.
Speaker 2 (01:11):
Is looking all right.
Speaker 3 (01:13):
But in all, in all seriousness, let's let's just dive
into the facts. Two plus years ago, we came out
with a really exciting manuscript showing that we were really
the first company to put out work in antibody Danovo
design where we could design the h CDR three region,
the most variable region of the antibody, to bind to
(01:35):
a particular target of interest. Again, this was the first
time anybody had shown any work in antibody Danovo design.
And you know, two years later, we're at the point
now where we can actually use these models to go
after really hard challenging targets where there is no reference molecule.
(01:56):
And a perfect example of this, which actually was not
covered in the STAT article even though we had talked
to the reporter about this was the work that got
unveiled at our R and D day in December, and
this was work in collaboration with Caltech doctor Steve Mayo
(02:16):
and doctor Pamela Workman, and they had identified a region
in the HIV virus called the Coldera region. This is
a highly conserved region and today it's been very difficult
to drug. But the belief is if you can drug
this Coldaa region, you could actually potentially create a neutralizing
(02:37):
antibody to all the different clades or variants of HIV.
And we were able to indeed use our model, our
Denovo model, to design an antibody to this coldera region
completely from scratch. There was no reference molecule there. We
(02:57):
designed all six CDRs and UH, this UH I think
really highlights UH where we want to be with Denovo design,
being able to unlock a new novel biology by being
able to go after and drug these very hard and
challenging targets, just like we did with the HIV coldair region.
(03:20):
Actually have two other examples of being able to do this, UH,
our partnership with Omorole going after an ion channel and
then with with astrozeneca, a very challenging UH target as
as as well. UH. And So no matter what definition
of Denovo you use, UH, the work we did with
cal Tech UH meets any definition of Denovo that that
(03:46):
you could potentially have. And we're really proud of that work.
And so at the end of the day, UH, the
work speaks for itself. Put it out there, UH and
UH and and and it's really exciting and and at
the end of the day, where we're at with the
Denovo design of anti bodies fits really well into our
overall corporate strategy moving forward, which is to go after
(04:11):
targets that have known biology but have been difficult to
drug and this, you know, we believe is going to
ultimately give us an advantage being able to create you know,
first in class and best in class assets. Yeah.
Speaker 1 (04:23):
Absolutely, that's it makes a lot of sense, and that's helpful. So,
I mean, I think just in general, biologics are typically
viewed as a simpler construct than small molecules. Given you
don't have as much rug, you don't have any off
target interaction, you don't have to worry so much about
the metabolism of it.
Speaker 2 (04:37):
Do you think I have some mischaracterization?
Speaker 1 (04:39):
And then kind of what variable is do you think
you can tune in a biologic and how does it
impact the outcomes?
Speaker 4 (04:44):
We think biologics are actually not a simple construct. If
you look at, for example, sequence diversity, just the combinatorial
space for biologics, that's actually larger by most estimates than
the possible space for small molecules. Although it's very quick too.
You're going to be very quick to reach very high,
very large numbers very quickly. But if you look at
(05:06):
a monoclonal antibody, which is the you know, the workhorse
in biologics, you know, it's one hundred and sixty kilodoten protein.
It's a pretty complex protein. So the combinatorial space in
biologics is pretty large, and that's also where the strength
of generative AI comes in. So the ability to really
(05:27):
sample this sequence space close to exhaustion is really the
opportunity we have using generative AI.
Speaker 1 (05:35):
Absolutely, and then maybe I think apps that refers to
their biologics pipeline as smart biologic So what makes these
smart in your view?
Speaker 4 (05:44):
Yeah, absolutely, if you think about you know, the need
in biologics. To date, Essentially the large majority of biologics
have been generated using either mouse immunizations or display technologies
such as phage display. These technologies traditionally are always limited
by so called immunodominant epitopes in the case of imanizations,
(06:07):
or by whatever is generated using a display technique. So
the ability to design smart biologics is on the one hand,
the opportunity to actually consciously select an epitope that displays
the pharmacology of interest. And then on the other hand,
you can go further. You can engineer multi valency, you
(06:30):
can engineer conditional binding for example, pH dependent binding. You
can engineer certain multi specificities that are quite interesting. And finally,
obviously you can you can enhance the mode of action
by tuning potency and tuning affinities and then finally also
(06:51):
tuning functional activities.
Speaker 1 (06:54):
And you touched on the conscious selection of epitopes there.
So how do you prepare your approach to maybe some
of your competitors they want to do everything soup to
nuts fully automated in their AI models, because I believe
apps I basically uses drug hunters to decide their epitopes.
Maybe you can speak to that.
Speaker 3 (07:10):
Yeah, at the end of the day, like there is
still a human intervention that that's needed. Uh. And and
that's because we don't have the data to go full
end to end at at at.
Speaker 2 (07:21):
The current moment.
Speaker 3 (07:22):
If you look at just kind of the history of
of ABSI, we we started off as a as a
data first company. We figured out how to scale protein
protein interactions, you know, essentially the drug functionality, how the
antibody interacts with the target of interests. We're able to
scale that from thousands of interactions to to two millions
UH to ultimately use this data along with other structural
(07:45):
based data, to ultimately build out our Denovo model where
you can you know, design an antibody from from scratch,
but you have to have that that data to you know,
ultimately build these overall models. And we're just not at
the point right now where there's enough you know, functional
(08:05):
data out there to be able to have an AI
model be able to predict the epitope that's going to
give you the functionality.
Speaker 2 (08:14):
Now, ultimately, that's the direction.
Speaker 3 (08:15):
That ABSIT is wanting to go, is is instead of
a drug hunter or a structural biologist picking the target
of interest, we can actually have the model start to
to predict that and and and those are some of
the uh, you know capabilities we're building out right now.
Is is how do you you know, scale the overall functionality?
(08:37):
Uh so you can actually use the models to predict
the epitopes in the future.
Speaker 2 (08:42):
But right now we're we're using the AI models.
Speaker 3 (08:45):
To design the antibodies to the epiitopes that we think
are going to have the biggest impact on functionality. But
again we're we're not in this world right now. We're
design everything with a with a click of a button.
I think you know, in five plus years right now
we will get there. But there is that that human
intervention that is absolutely critical to the success of of
(09:08):
AI uh for for you know, for us and I
think others.
Speaker 2 (09:13):
In this in this field as well.
Speaker 1 (09:14):
And that's the data for the epiitope side. But let's
talk about the data for the antibody sequence side. I
think some of your competitors have talked about using public
data sets and the prominence of maps without mutations because
they're derived from naive B cells, so that reduces the
predictive ability for any sort of German mutation you introduce
into a sequence.
Speaker 2 (09:33):
How has apps I worked to overcome this.
Speaker 4 (09:35):
Yeah, you're totally right, Andrew. The scarcity of data generally
in the antibody space is the limitation because obviously you
need to train your models to really be able to
journalize across different target domains. And you know, beyond the
mutations you just mentioned, we think actually the bigger the
(09:56):
bigger issue is the lack of structural data in a
public domain. The public domain right now has around seven
hundred structures of antibodies with their cognate antigen So that's
clearly magnitude a way of what would be required to
have a fully generalizable model, that is that can be
(10:17):
meaningfully trained. But that said, there are ways to address
this obviously, you know, we have been very conscious in
generating and acquiring data sets in addition to the publicly
available data sets to enhance our models. And then we
also use additional technologies, for example, more likectodynamics to first
(10:39):
further diversify and validate our data sets. So I think
it's at the moment it's very much about being being
smart and very very conscious about what the data currently
entails and how to enhance it. And I think in
the future we'll see a lot more data becoming available.
But at outside we view data as finally as as
(11:01):
the final mode of being successful using generative AI, and
that's what we've building been building over the years. We've
been building a large set of affinity sequence data pairs
that ultimately allow us to really uh improve our model significantly.
Speaker 3 (11:21):
Yeah, and and just to kind of you know, hit
on that a little bit more, I mean, you know,
not only are we using your publicly available structure data,
we we have our obviously our sequence data where we're
generating you know, in house. And I think a direction
we're actually really excited about and in the future that
that Christian hit on is the the MD simulation and
actually using MD stimulation to generate synthetic data as as
(11:45):
as well to to enhance the overall generalizability of of
the models. And and so you can see that, like again,
data is so key to getting to the point of
generalizability within biology, and that's has been a huge focus
for us, whether that's experimental data or generating synthetic data
(12:06):
through MD simulation.
Speaker 1 (12:09):
A broad philosophical question almost do you think smaller bio
tech biostartups are kind of a disadvantage on the data
side versus these kind of large cap pharma companies that
have decades of development experience and even if all their
data is untagged, you would assume it's somewhere there in
the background.
Speaker 3 (12:26):
Yeah, I can hit on this and then you know,
Christian can you know, speak to his his large pharmat experience.
But I think what ends up happening in large pharma
is that, yes, they generate a ton.
Speaker 2 (12:36):
Of data, but it's very diverse data.
Speaker 3 (12:40):
It's it's you know, there's not you know, you know
a lot of you know, let's say, structural based data
on a particular problem you you want to solve, and
so you know, to solve a you know, a particular problem.
Like Denovo design, you have to have a lot of
a you know, specific kind of of of data. And
I would say that, you know, pharma again doesn't have
(13:03):
you know, a you know, in some cases they have
a lot of you know, specific data, but it's it's
more broad, uh, data across you know, many different types
of of data.
Speaker 2 (13:15):
And so there's only so much that.
Speaker 3 (13:17):
You can you can ultimately, uh, you know, leverage that
and and use it for And I think you're seeing
pharma do that and and in the areas that they
don't have the data and they're not getting the generalizability
on the models.
Speaker 2 (13:29):
You know.
Speaker 3 (13:29):
I think you're seeing them you know, partner with with
biotech companies, uh to to be able to kind of
build out those those capabilities. But uh, I mean, uh
a Christian please step in with the you know, your
experience at at at at buyer.
Speaker 4 (13:43):
Yeah, thanks Sean. Now, just to add it to what
Sean said, I spent more than a decade in big Farmer,
so I think I know what what how this looks
from the inside. I think you're not going to be
surprised to hear that there is, indeed a lot of
data available in big pharma and on paper that should
give them an advantage, but the reality is that usually
(14:06):
that data is not accessible, and it's not available in
a format that actually allows machine learning models to be trained.
So most of the big farmers actually going through a
massive digitization and transforming digital transformation process to make this
data available and actually make those available for machine learning models.
That said, I think the overall magnetude magnetitude problem that
(14:29):
I alluded to earlier remains right, because obviously the amount
of data each single farmer company has internally is still
a fraction of what's available publicly. So I think even
if they're successful with that transformation effort, the overall, the
overall amount of data is still is still limited. So
(14:51):
I think you know to your philosophical question earlier. I
think in the end, what we decided to do is
read to be very cognizant about what we want to
generate in terms of data that is really additive to
an genitive AI model, and then be very thoughtful about
how can we complement that with synthetic data, for example,
(15:12):
using molecular dynamics. We think that actually puts us in
a bit of an advantage because we are very quick
to it to rate and we're very thoughtful about how
to make the best out of these models. So I
think it's a it's an exciting race to be in.
I don't think we're at this advantage being a small
company at all.
Speaker 3 (15:32):
Yeah, And I think one thing in addition to being
a small company and having kind of the data platform
that we have is it's allowed us to create this
this lab in the loop process where we can rapidly
go from data generation in the wet lab to training
our models to then go in and validating those and
we can do this in a six week time period,
(15:52):
so we can you know, almost like a tech company,
rapidly iterate on the model architectures, you know, the designs,
the hyper parameters to really kind of figure out the
ultimately the best data sets to be pairing with the
best moldor architectures. And this has given us a huge advantage.
And I think that that's another reason, this lab in
the loop why we've in two years been able to
(16:13):
go from you know, designing you know, uh an h
cd R three of an antibody to designing now all
six CDRs to challenging targets. Because we've been able to
have this lab in the lab in the loop, and
I think that that is very differentiated and and you know,
being nimble and fast and having that type of technology,
(16:33):
I think is you know, something that is usually associated
with a you know, smaller, more nimble company versus a
large uh you know, big pharma.
Speaker 1 (16:43):
And you touched on this proprietary wet lab that you
have as part of your lab in a loop. So
I guess maybe do you want to discuss how you
think it's differentially and how it fits into this kind
of iterative learning process and providing reinforcement maybe much faster
than other AI enabled companies would.
Speaker 2 (17:00):
Yeah. Absolutely.
Speaker 3 (17:01):
So we're you know, we're able to ultimately, uh you know,
screen the these protein protein interactions. Uh, you know, in
a very uh you know, fast and effective manner essentially
how antibodies are are interacting with with targets of interest.
Speaker 2 (17:17):
Uh.
Speaker 3 (17:17):
And you know we can you know, we can scale
this up to to generate you know, millions of interactions
to you know, for for training. But another key piece
of this this technology is that you're able to actually.
Speaker 2 (17:29):
Use it for for validation.
Speaker 3 (17:31):
Uh, so you can actually see how accurate your your
your models are uh and and go and test, you know,
hundreds of thousands of uh discrete, UH specific sequences that
that are coming out of of your model. Again to
to see how accurate your your your models are. And
so we're using it both on on the training side,
(17:54):
but also importantly on.
Speaker 2 (17:56):
The overall validation side.
Speaker 1 (18:00):
Maybe just I know they're probably not statistic behind this,
but has that affect the actual outcomes and the I
guess closeness of the model to the actual outcome relative
to some companies that might just do purely in silica
reinforcement learning for their models.
Speaker 4 (18:15):
Yeah, look, I mean it's it's difficult to compare us
to others because you know, we don't know what what
hit rates others have. We've published our you know, in
some of our pre prints, we've published our hit rates,
and I think they are competitive and they are they
are in particular when it comes to targets for which
(18:35):
classical technologies have have failed, I think they are quite remarkable.
So again, coming back to the Caldera example that Jean
mentioned on the HIV Buyers, I think that's something that
has never been achieved before, and I think that's a
testament to the fact that you know, this technology allows
us to really address unmathematical need in a way that
(18:59):
has not been possible before. So for us, you know, AI, Gennifer,
it's not just a tool we want to use because
it's fun and it's you know, it's fancy, but it's
really the opportunity to deploy it in a way that
you know, addresses un mathematical needs.
Speaker 1 (19:17):
Okay, And then I think when we talk about small
molecule companies with AI, they talk about the chemical space
ten to the sixtieth, ten to the twenty third whatever
their estimate is, and how they want to explore the
unexplored chemical space. What do you think the analog is
for biologics? Is it just antibody sequences or maybe you
can take it from there.
Speaker 4 (19:36):
Yeah, sure, I mean the classical way a protein engineer
views the combinatorial space and an antibody is you essentially
take the complementarity determining regions of an antibody and you
essentially take the entire sequence space that's possible in these
six CDRs that a human antibody has, and then you
(19:58):
permute all of those.
Speaker 2 (19:59):
Right.
Speaker 4 (20:00):
Obviously, the human immune system has limitations it can it
typically does not permute all those in the computer we
can do that. Obviously, we need to pay attention that
we stay human humans is critical because because we don't
want to a generator molecule that is immunogenic, has toxicity liabilities,
has the propensity to form anti drug antibodies. But within
(20:22):
that boundary, and our models are obviously trained on human sequences.
Within these boundaries, the addressable sequences space is enormous and
I think that's the that's the opportunity. So depending on
how you look at it, you could you know, ballpark
the sum of all ammune acids across all CDRs is
(20:44):
probably anywhere between fifty five and seventy and then you
take that number to the power of twenty, which is
then the number of amino acids you have, So that
gives you roughly the combinatorial space you need to substract
that those sequences that are not human, right, that would
be sort of a back of the envelope calculation.
Speaker 3 (21:06):
Yeah, and I and I think like one of the
advantages to that that we have being in the you know,
in in the biologics space is that we I guess it's.
Speaker 2 (21:17):
A plus animus. You have you know, living organisms.
Speaker 3 (21:20):
That that have to actually you know, produce the protein
of interest versus a you know, medicinal chemist, and so,
you know, I think some of the issues on the
small molecule.
Speaker 2 (21:29):
Side that that that arise is that.
Speaker 3 (21:31):
You can you know design, you know, use generative AI
to design these really exciting you know small molecules, but
they just may not be you know, possible to to
actually make. And it's very time consuming to to have
a chemist go and synthesize you know, all of these
very you know, specific chemistries and hits that the model
generates versus you know, we can easily again using our
(21:54):
own wet lab technology easily you know, build out you
know one hundred thousand and uh you know sequences that
the model produced and and have you know our you know,
our platform manufacture those and and go and test and
and screen those again in kind of that six week
uh you know time period, and so we can kind
of make that that rapid uh iteration I think much
(22:17):
faster than the you know, small molecule folks can can do.
So I think there's a bit more flexibility, uh and
and ability to to iterate uh and and and and
go kind of through these you know lab and the
loop processes you know, much more with biologics than small molecules.
Speaker 1 (22:38):
And then I want to get into your individual assets,
but maybe just one high level question. When you think
about AI's potential to change the drug development paradigm, is
it time and cost and the pre clinical stages or
do we think that you can change the probability success
for drug development more broadly?
Speaker 3 (22:52):
Absolutely, Look, we're we're focused on solving uh, hard problems
that that that still exist. I think, yes, it's it's
great to do things is you know, faster and cheaper,
but that's not why we built out our r AI platform.
We we really built it out to solve the challenging
problems that still exist with within biology going after these
(23:13):
these undruggable UH targets. And and by being able to
go after those undruggable targets, you you have the ability
to now address uh, you know, diseases and indications that
you know previously drugged or or or solved.
Speaker 2 (23:28):
And so I think that this is why we we do.
Speaker 3 (23:31):
What what what we're doing, and and you know, ultimately
we want to utilize this to create you know, first
and class, best in class assets and and ultimately, yes,
increase overall probability of of of success and and uh,
you know open up the the you know, the different
diseases that that are possible to ultimately solve.
Speaker 1 (23:51):
Okay, great, So let's talk about a b S one
on one, your lead wholly owned asset t O one A.
So you're coming from behind against some large editors to Barley.
Where do you see the holes in the current product
profile from your rivals?
Speaker 4 (24:05):
Sure? So what we what we set out to address
with ABUS one on one when we started the project
eighteen months ago, as we looked at what's already out there,
there are three competitors. There were three competitive molecules, as
you mentioned, and all of them had liabilities. Some of
them has have liabilities in terms of feminogenicity, which means
(24:25):
a high propensity of a patient developing anti drug antibodies,
which may then lead to a lack of efficacy in
these of these drugs, over insufficient pharmacokinetics or requirement to
dose more frequently a requirement of so called induction dose,
(24:46):
which is intravenious. And then finally there's also a question
regarding efficacy and the ability to dose higher. So we
looked at all these parameters and we define a target
product profile for ABS one oh one that addresses all
of these shortcomings. So in essence, what we've been able
to achieve with ABS one one is a more potent,
(25:09):
high affinity molecule that has very good developability, which means
it can be formulated a turn a milligrams permil dad
and in fact we already achieved that, So even before
going into first in human we have already now achieved
a turn a milligram per formulation, which is usually what
what others do much later in the clinical development plan.
(25:32):
We've also addressed the umunugenicity issue. We have pre clinical
data that demonstrate that we have much less of such liabilities.
And then we also engineered half life, which means our
antibody is capable of achieving a much longer presence in
the plasma, potentially allowing every two to three months dosing.
Speaker 3 (25:54):
I will not too in the NHP studies, we are
also seeing the potential for you know, better biodistribution and
and so we believe with that we could also potentially
eliminate a loading doses as well, which you see with
some of the other competitor molecules.
Speaker 1 (26:11):
And it's one thing about your designs that a BS
one oh one has a dual affinity for both the
TL one A trimers and the monomers, whereas most of
your competitors focus more so on the trimer. So why
do you view monomer activity as important?
Speaker 4 (26:24):
Yeah? Absolutely, we get that question a lot, and it's
it's interesting, U. There's actually published literature out there that
that seems to be somewhat overlooked that clearly shows that
there are isoforms and fragments of tier one A that
remain monomeric through protioli protiolotic cleavage, and those can no
(26:45):
longer form traumeric structures. However, these short of fragments have
been shown to elicit subtle effects. In other words, they
can induce apoptosis and they can stimulate into fer comma release.
So we think it's reasonable to assume that in patients
with increased levels of trimeric to one day, there will
be also an increase in monomeric to one day through
(27:06):
these shorter isoforms. And we believe that having the ability
to bind the monomer as well as a trimer could
pretend potentially result in a treatment benefit.
Speaker 1 (27:21):
Then you touched on the pre clinical experiment that showed
low complex internalization, which should drive lower rates of antibody
anti drug antibodies and less imunogenicity. But as that panned
out in clinical data, we've seen the date for your competitors, we.
Speaker 4 (27:35):
Can only talk about what the competitors have disclosed. So
for the ROSI molecules, we've seen in a phase one
study that fifty six out of sixty eight patients have
developed anti drug antibodies, and that's the molecule where we
think we see also an increase in internalization. So we
think there is a pretty good correlation between that internalization
(27:59):
assay and in the end clinical ada rates. It's interesting
that's actually not only the case for for Tier one A. UH.
There's a more historical example of this effect out there,
which is on tinafalfa. There's a there's a molecule called
inflix mapp inflex the map for which this effects has
been described for the first time. So the inflex mapp
(28:22):
immune complexes are also quite.
Speaker 3 (28:27):
UH.
Speaker 4 (28:28):
They have been described to elicit antidoch antibody formation as well,
so there's precedence and we think the same effect you
see an inflexing map will apply to a Tier one A,
which is why we have selected an epitope that has
much less of this propensity to drive immune complex driveniminotenicity.
Speaker 3 (28:48):
Yeah, and kind of going back to to the AI
platform and uh, you know, we we utilize the AI
platform to design the antibody.
Speaker 2 (28:57):
To a particular epitope that we thought was going to.
Speaker 3 (29:00):
The immune privileged and and you know, ultimately wasn't going
to elicit uh, you know, a B cell.
Speaker 2 (29:07):
Response. Uh.
Speaker 3 (29:08):
You know, based on the clinical data you saw out there,
you know, MERKS eighty A rate was was was much
lower than than the roy VAN and if you looked
at just the T cell activation of of those, you know,
they they were very similar, which again led us to
believe that this was B cell mediated. And so we
designed an epitope you know, closer to the MRK epitope
(29:31):
because we we thought that that was an immune privileged
site that you know, ultimately would would elicit a much lower.
Speaker 2 (29:40):
B cell response.
Speaker 3 (29:42):
And again, so that's kind of a way that we
utilized our AI platform to engineer in uh, you know,
a specific attribute in this case, you know, potentially lower immutagenicity.
Speaker 1 (29:56):
They made another question about epitope choice. So t O
and A as aand has two receptors, DR three and
then d c R three or decoy receptor three. So
your competitors have talked about preserving binding of t O
one a to d c R three as basically a
sink of free t O one a. Do you think
that's important for ABS one O one and does it
do that?
Speaker 4 (30:15):
Yeah, very good question. So we have looked at d
R three and DCR three binding, and we have also
obviously tested all known antibodies out there, and we essentially
determined the racials of d c R three and DR
three binding. We think ABS one on one has demonstrated
(30:37):
the best ratio of DR three tow DCR three binding,
so the best selectivity profile at least in our hands,
and we think this could be a differentiating feature. On
the other hand, it's in a way arguable because the
DCR three concentrations inpatient in patients is actually very very low.
It's one thousand and four lower than the doses at
(31:00):
which the monoclonal antibodies are administered. So arguably, if there's
a sync effect, it's probably negligible. But if there's a
sink effect and it's meaningful, then ABS one oh one
has actually the best selectivity ratio.
Speaker 1 (31:13):
Then maybe maybe past the antibody design and thinking forward,
some of the competitors have adopted biomarker led approaches for
proof of concept trials, whether that single nucleotide polymorphisms or
soluble too one A levels. Does AP think that that
you're going to take a similar tact and how important
do you think identifying these potential super responders are?
Speaker 4 (31:32):
Yeah, obviously that's a super interesting field. We've watched it
very closely. You have probably seen the recent data published
around to the soakeybart in the Artemis you see study,
and in that study they showed no difference in outcomes.
In other words, an enrichment at least using the similar
(31:55):
nuclear tide polymorphis since they have selected in that study
did not result in any and in any treatment benefit.
So it's it's at this point I think, still a
valid idea, but it's actually very difficult to put it
in practice. The other piece that that is still you know,
out there for for the market to to to decide
(32:15):
is how how is the uptake going to be with physicians? Right, so,
to what degree will a physician based treatment decision on
such a test? I think an all comer antibody actually
has a big, big advantage there because if you are
able to demonstrate very good afflicacy and an all commer approach,
(32:38):
and I think it is one one is is very
well positioned to deliver such a target product profile. You
know that you can argue that you could those even
higher to to deliver a better efficacy. So, in other words,
I think the the all comer approach is probably going
to be the one that's going to be more attractive
(33:00):
and more successful in the market.
Speaker 2 (33:05):
Yeah.
Speaker 1 (33:05):
Absolutely, So we talked about aside AI platform for antibody
design and everything else a lot about the companies we've
talked about using AI in their clinical trial conduct and design.
Does Abside plan to start integrating that as well as
you move into the clinic.
Speaker 4 (33:20):
Yeah, Look, I mean this is a really exciting field. Certainly,
the entire value chain of research and development and the
pharmacy real industry is ripe for disruption by artificial intelligence.
We are exploring actively how we can use genitive AI
in such processes as well. I think it's fair to
(33:41):
say this is not our focused area right now. We
are laser focused on delivering biologics using genitive AI. But
we are certainly exploring this, and I do think we
will see a lot of innovation in these fields going forward. Yeah,
and I think it will span the entire value chain,
starting from patient selection all the way through regulatory submissions
(34:04):
and clinical trial protocols. So there will be innovations in
the field and we actively exploring those as well.
Speaker 2 (34:11):
Yeah.
Speaker 3 (34:12):
Look, as as Christian said, you know, we're we're not
pursuing this ourselves, but you know, these are you know,
areas that you know, we we see as as being
you know, you know, hugely transformational. And I think these
are areas definitely that you know, we see ourselves, you know,
partnering you know, with with leading AI clinical development you know,
companies in the future. But I think you know, given
(34:35):
the market you know, dynamics, and and the need to
you know focus uh, you know, we've decided to focus
in on on the Denovo design of antibodies. But again
I think, uh, you know, AI is transforming across the board,
and we're looking to partner uh with with folks that
are you know, you know, partner or that are utilizing
(34:55):
AI not only in in uh you know, clinical development,
but also you know, potentially target discovery and anywhere we
can integrate it to ultimately, you know, get the best
outcomes possible.
Speaker 1 (35:10):
Maybe can you just give a quick status update on
ABS one on one I woul understand you're still an
I in d enabling studies.
Speaker 2 (35:15):
When do you plan to enter the clinic?
Speaker 3 (35:17):
Yeah, so we're going to be entering that the clinic
here within the next couple of months, which is really exciting.
And we plan to have a Phase one interim readout
in the.
Speaker 2 (35:31):
Second half of this year, in.
Speaker 1 (35:34):
Which we expect from an interim readout given a Phase
one will probably be in healthy volunteers.
Speaker 4 (35:40):
Yeah, so what we will read out and this is
probably not surprising, it has been well described for Tier
one A. You can measure solubil Tier one A in
plasma and in healthy volunteers you can expect an elevation
of solubility one A when it's liberated from the tissue
bound by the antibody and then getting back into the
sexual relation. That's a very reliable and de risking biomarker.
(36:05):
It's a target engagement biomarker, and in particular for our
extended half life proposition, we hope to see a sustained
elevation of solubility one day in healthy volunteers after a
single DOLLS. We think that's going to be a significant
d risking of the program overall.
Speaker 3 (36:22):
Yeah, and you know, we we recently at JP Morgan
presented the target engagement data that we were able to
obtain from our non human brainmate studies and and that
was a you know, a really beautiful profile of of
you know, our our target engagement compared to the the
competitor molecules. And and that you know profile is a
(36:46):
similar profile that we're hoping to to see in uh
IN in our phase one interm readout.
Speaker 1 (36:53):
As Christian was mentioning, and maybe after you think on
the other side of this interim analysis for the phase
one and as you move in to prove a con
axcept an individual indications. IBD is obviously kind of the
classical TiO one A indication, but we see now that
some of these companies are starting to expand into other indications,
other inflammatory conditions. So how aggressive is appside planned to
(37:14):
be an indication to both in IBD and also outside IBD.
Speaker 4 (37:18):
Yeah, So TILL one A is a super exciting mechanism
because it has a lot of potential outside of IBD,
and we are absolutely looking at indications outside of IBD
as well. We have not this closed publicly what our
plans are specifically, but I can assure you that beyond IBD,
(37:39):
I think the market potential will be significant and hence
it would be improvedent not to explore it. So that's
what we're doing. In the end, we will go a
route that allows us to derisk the program as fast
as possible and then allow us to move into late
stage clinical trials in an indication where this more think
(38:00):
you can deliver the highest value.
Speaker 3 (38:02):
Yeah, and I think it is you know, one thing
you know, I will note I think it is important
to be able to to show potential, uh you know,
superiority to the clinical competitors, and so you know, definitely
at least you know, doing a a you know, proof
of mechanism uh in in UC uh to show you know,
how maybe S one on one compares to the clinical competitors.
Speaker 2 (38:26):
I think it is obviously really important.
Speaker 3 (38:29):
And then as Christians that we're looking to see, you know,
what are some other uh you know, indications we could
go after, because I think that there are some really
exciting opportunities to to to go after there.
Speaker 1 (38:41):
Maybe you can shift gears now and talk about ABS
two one, which you recently unveiled at your R and
D day in December. Maybe just walk us through the
genesis of this program because biologics for what are broadly
used as cosmetic indications haven't really been seen in the past.
Speaker 2 (38:57):
Yeah.
Speaker 3 (38:58):
Absolutely, you know, maybe just before we uh close out
one on one, there's one additional point that I that
I wanted to make on on one on one. Uh.
You know, if if you look at the duet you
know trial that janej is doing, it's a it's a
combo based you know trial in ib D.
Speaker 2 (39:14):
Uh. And and we really.
Speaker 3 (39:16):
Do you know, believe that you know, combo based or
or a bispecific approach is ultimately going to be the
way of the future.
Speaker 2 (39:24):
Uh.
Speaker 3 (39:24):
And you know, one of the things we're actually really
excited about that we've been developing is actually a Tier
one a bi specific and and this by specific is
not the the other arm is is not a known
target or at least a target that's been in discussion.
It's not an Aisle twenty three or an Alpha four
(39:46):
to seven.
Speaker 4 (39:47):
Uh.
Speaker 3 (39:47):
This is actually uh, you know, the other arm is
is actually a known I and I target that has
kind of been difficult to drug for for various reasons.
And you know, we think that this as as a
follow on to our our best in class or potential
best in class Tier one A and A body is
(40:09):
you know, potentially going to I think be a you know,
really exciting asset to follow.
Speaker 2 (40:15):
And and so.
Speaker 3 (40:16):
Anyways, wanted to cover that last little piece on one
on one before diving into two one.
Speaker 1 (40:21):
Yeah, I appreciate you throwing that in.
Speaker 4 (40:23):
Yeah, yeah, I can cover two one, So you're totally right, Andrew.
I mean, the aesthetics field is typically a very risk
averse field. That's why you have not seen a lot
of innovation there. Actually, the last novel treatment for androgenic
alopecia has been approved more than twenty years ago, So
this field has has a scarcity of innovation, which is
(40:46):
why we're very excited about it is two to one.
We think the market potential for energen alopecia is enormous
and it's actually growing. So ABES two to one really
has the potential to not only address hair loss, it's
actually a mechanism that potentially allows hair regrowth and also
(41:08):
address pigmentation problems. So based on the data we see
pretty clinically there could be an opportunity to actually get
back your natural hair color as well.
Speaker 3 (41:19):
Yeah, I mean if you talk to to dermatologists, you
talk to to patients that they're looking for a hair
regrowth uh product or medicine that can that can truly
you know, regrow their hair and and have a durable effect.
I think a lot of the medications that are out
(41:41):
there are are are hit or miss, and so if
you can you know, deliver to to you know, patients
a you know, a drug that can give them hair
regrowth as as well as the durability of of that,
I think that that, you know, that's something that that
patients i think are really yearning for. And as as
(42:05):
Christian said, it's it's a massive opportunity.
Speaker 2 (42:08):
I mean, eighteen to ninety million.
Speaker 3 (42:10):
Americans suffer from androgenetic alopecia and and you know, the
last you know medication that or the lestra that that
was reproved was you know, twenty five plus years ago.
Uh And and so they're they're they're you know, in
terms of the the competitive landscape, there's there's there's not
much out there.
Speaker 2 (42:29):
It's a big opportunity.
Speaker 3 (42:31):
Uh, and we see ABS two to one as as
being you know, hugely transformational here.
Speaker 1 (42:38):
And maybe just touch on your choice of target prolaction receptor.
So there is a relationship and a shared pathway between
the prolaction receptor and JACK stat pathways. So what level
of confidence do you have in prolactin receptor given JACK
and hibbers have had sidely mixed results in androgenic alopecia
compared to other forms of alopecia like alopecia areata.
Speaker 4 (42:59):
Yeah, yeah, yeah, absolutely. So first of all, it's I
think important that we distinguish these two different different diseases.
We have energenic alopecia, which is pattern hair loss and
that's really driven by androgen signaling, so it's a hormonal
path physiology. And then there's alopecia areata, which is an
audimmune disease that results in hair loss. And it's correct
(43:23):
that JACK inhibitors have shown mixed results in energenic alopecia
and they also have a problematic safety profile. So these
kindness inhibitors have black box warnings. Some of them have
been approved in alopecia areata, but clearly those are systemic
(43:46):
JACK inhibitors, so they will also inhibit the JACK signaling
in your immune cells which results in im you know, suppression,
and this obviously results in side effects such as urinary
tract infections, pulmonary infections, and other problems which actually resulted
in these Black Books warning. So we think protactin receptor
(44:07):
modulation is distinct from the JACK inhibitors. And the prolactin
receptor is first of all, largely specific at least addressable
with a biologic largely specific to the half vollicles. And
the second of all, it has a dual role. It's
not only acting through jacktool stigaling, but it's also acting
(44:29):
through STAT five signing. So there is clearly a differentiation
to JACK inhibitors.
Speaker 1 (44:38):
And you touched on the safety profile of JACK hibbers,
which is well known and well described. But I think
we have some other prolactin receptor. In a BIS, you
start to see increased in the serum prolactin due to
the inhibition of the negative feedback on petuitary levels. Has
absite considered a way to circumvent this and what risks
do you think could be carried with increasing prolactin levels.
Speaker 4 (44:58):
Yeah. So look, we have very we are very convinced
about the safety of targeting the productin receptor. There's human
genetic evidence that suggests that targeting the producting receptor is
very safe. There's a report out by by researcher described
a female individual who has a complete lack of lack
(45:19):
of productin receptor signaling and she's healthy UH. And note
that this is a lifelong human knockout. She's given birth
to two healthy children, and she has overall normal serum
electrolyte and hormone hormone levels, so no abnormalities, with the
exception that she was unable to lactate, which is not
(45:40):
surprising if you target prolactin. So that suggests that indeed,
in that individual there was an elevated productin UH level observed.
But that suggests that that elevation of prolacting in and
of itself is not a concern. So the other piece
you have to consider, we are are certainly not a
(46:02):
lifelong aiming to deliver lifelong treatment. And also you have
to you have to keep in mind that about logic
does not for the most part, past the blood brain barrier.
So the ability opportunity is that we are not to
(46:22):
a large extent not hitting the pituitary. We're not hitting
the central norvil system. The antibody will go into the
half vnicles and that's where the action will be.
Speaker 1 (46:33):
And I know you haven't shared too much about a
b S two one profile yet, but where do you
see the potential to improve on realistically? The one competitor
in the field right now the Buyer Hope product, So we.
Speaker 4 (46:44):
Are familiar with that molecule. We also know the liabilities
of that product. I think we try to address all
these liabilities with ABS two O one, and in essence,
what that allows us to do is really to deliver
an anti body that is highly manufactural and highly developable,
because in the end, we want to deliver a very
(47:06):
convenient product, which is a solution that is suitable for
subcutaneous administration, ideally a self administration in an auto injective
device that requires also solubility. We've shown very high solubility
of our antibody. We've also engineered half life extension in
that molecule, and through uh the half life extension, we
(47:30):
can also increase bioavailability, so increased organ exposure through an
increased transpseytosis effect. So if you take all of these
attributes together, I think ALS two to one is highly
differentiated to the existing antibody that's out there.
Speaker 3 (47:48):
Yeah, and the other issue too with that antibody is
that you know that the pad life on that is
pretty short. So obviously with what we have here, you
have an extent to patent life. And you know, given
that Hope is running this this trial in China, we
(48:10):
we do believe that we could be the the first
in the US market for approval, which we see as
another huge, huge benefit as as well.
Speaker 1 (48:24):
Is there anything else you want to mention about two
on one before we talk about HIV cluderra.
Speaker 3 (48:29):
Again, we see ABS two one as as our flagship asset.
You know, we we see in terms of overall market
size definitely greater than thirteen billion dollars a year. We've
been able to you know, to not only you know,
show you know, hair regrowth, but we we do believe
(48:49):
that this is going to have a durability. If you
look at you know, NHP studies that have looked at
anti prolactin antibodies, you know, after uh six months of treatment,
you see durability for four plus years. And and that's because
you're you're basically taking the follicle and putting it back
(49:10):
into the anagen or active growth state. And you know,
once you're in that state, based on your genetics, you
can stay there from from from anywhere from you know,
two to to six years. So that's you know why
we think we're going to get really strong durability.
Speaker 2 (49:25):
Uh.
Speaker 3 (49:26):
You know, you only have to dose for six months
and and you you have hair growth for the next
you know, two to six years. We see it as
a as a very safe uh you know product, uh.
Speaker 2 (49:35):
You know, based on the human genetics.
Speaker 3 (49:39):
Uh and uh and and and again it's it's a
it's a very large patient population that that we're serving here,
and so you know we're we're we're rushing to to
to get this into the clinic uh and and hoping
to you know, have some efficacy readouts on on that
next year.
Speaker 2 (49:57):
But again, we see this as a as.
Speaker 3 (49:58):
A really exciting asset and kind of our flagship outse
that's moving forward.
Speaker 1 (50:03):
So many HIV just leave it very open ended to
gauge your excitement around being ABS both as treatment, pre
pre exposure, prophylaxis, and maybe even curative in the sense
that they could potentially drive out viral reservoirs.
Speaker 4 (50:17):
Based on what we could show, we have something in
hand that has the potential to indeed be broadly neutralizing.
The region we targeted on the HIV virus is an
epitope known as the Caldera region, and this is a
highly conserved region. And if we are successful to actually
(50:40):
make this into an antibody that that works in a
broadly neutralizing way. We are running these experiments as we speak.
It could it could potentially be a very attractive broadly
neutralizing asset for for PA patients with HIV. So it's
something we are actively working on. We've we've an excellent
(51:02):
collaboration with our collaboratives at Caltech, as Sean mentioned, Steve
Mayo and Pamela Bjorkman, and it's it's a great privileged
and opportunity to be working on this with them.
Speaker 1 (51:12):
And then maybe just in our closing minutes, let's talk
about a few broad AI questions. One, where do you
think we're currently out on the Gardner hype cycle with
regards to AI and drug development?
Speaker 2 (51:24):
Uh?
Speaker 3 (51:25):
Yeah, I definitely think we're We're coming down from from
the overall hype.
Speaker 2 (51:31):
I think, uh, you know, we're we're I think getting.
Speaker 3 (51:34):
Close, not close to the trough, but maybe uh, you know,
not quite there. But I think we're we're probably going
to get to to the trough, uh, you know, probably
in the next one one to two years. And then
I think we're we're going to you know, start to
you know, come come.
Speaker 2 (51:47):
Out of that.
Speaker 3 (51:48):
But but at the end of the day, like this
is not you know, a fad that's gonna you know
ultimately you know, fizzle out. We we are actually seeing
how ai IS is transforming this this industry, not only
in what we're doing, you know, like going after you know,
the coldera region in HIV, going after these hard challenging
(52:12):
or un drug ble targets, but you know, we're we're
we're seeing big wins on you know, you know, clinical development,
on on.
Speaker 2 (52:17):
The target discovery side. I mean just just across.
Speaker 3 (52:20):
Uh the board of ai IS is is here to
to stay and uh and and I think it's going
to have a profound impact on not only that the
the you know, the R and D costs and time,
but ultimately probability of of of success. And it's ultimately
going to take uh you know time to you know
(52:41):
see uh you know these stats you know pan out
just due to the regulatory process. But you know, even
look at you know what we've been able to do
on on you know, the the speed and and and cost.
You know, we we were able to you know, get
to a drug candidate in ABS one on one in
in in fourteen months, uh, you know, and be able
(53:01):
to get in into the clinic for for roughly you know,
thirteen to fifteen million dollars. And you know that could
normally cost you know, a normal biotech or large pharma
you know, fifty to one hundred million dollars. And so
we're already seeing big savings there. We're able to see
you know, us go after undruggable targets and and so
I but you can see, I mean there's like the
(53:24):
skeptics that that are that are coming out and and
so again I think we're we're you know, we're coming
down from the from the overall hype as as seen
in the in the in the stat article. But at
the end of the day, AI is here to stay.
Speaker 1 (53:36):
And what are you most excited about what's coming next?
Both for the abside platform and the field more generally.
Speaker 4 (53:41):
What I'm really excited about is the opportunity to really
deploy AI in a way that allows us to deliver
assets that have been elusive, right so, really working on
previously undragable targets. I think this is the This is
the frontier we are tackling right now, and then the
next thing after that will probably be a way to
(54:03):
view the target space in a more combinatorial way. So,
for example, if you look at one of the most
recent successes in the biologic space was an antibody called
the mechism ap, it's actually targeting into Luke in seventeen
A and into Luke in seventeen a f so targeting
two iso forms of an interluken and they could show
that this antibody is superior to what's already out there,
(54:25):
So dual inhibition in that case made sense. Now you
can ask the question why target only two or why
target these two? And I think the opportunity to actually
view the target space in a way beyond one seeing
a target is super exciting, and biologics using generative II,
there will be biologics that can do this, and this
(54:49):
is I think the next frontier that we'll see.
Speaker 3 (54:51):
After that and you know, we live in a really
exciting you know time time right now.
Speaker 2 (54:56):
I mean you look at you know what's going on
in an AI.
Speaker 3 (54:59):
You've had the the deep zeek uh moment and UH
and and you know we've we've kind of experienced in
biotech a little bit of kind of this you know,
deep seek moment or ourselves. I mean over this past year,
you you look at, you know, all all the acquisitions
and licensing deals that are you know, happening from from
from China. They're they're not licensing assets from uh, you
(55:21):
know US, you know biotech companies. You know, I think
it was over thirty thirty percent plus UH you know
assets over the over the past year, I've been you know,
either you know licensed UH or or bought from from
Chinese companies. And and that you know really kind of
you know bigs. A question like how you know, as
as US you know biotech companies, how do we continue
(55:43):
to you know, differentiate UH and and you know create
you know, novel breakthrough UH assets and and I think
it's it's leveraging AI. It's not to do it you know,
faster and cheaper and to compete with with with China.
It's it's actually to do something completely novel, uh being
able to again go after you know, uh, you know,
(56:05):
hard to drug targets like the cold Are region or
even like the by specific that we're we're talking about
where it's like, you know, you have you know, agonism
at higher concentrations when you want antagonism, how can you
you know, eliminate that agonism and and focus you know,
solely on the antagonism uh and and create you know,
these differentiated profiles like you know, pH dependency or multi
(56:26):
valency to ultimately increase overall probability of of of success.
And and that's really again where we're focused in on.
It's not better and faster, it's it's it's how do
we utilize uh ai to to to solve these these challenging,
you know, problems. And that's uh, I think, uh, you know,
the future for for ab sign, I think how uh
(56:46):
you know, we're going to see the apps you know,
AI differentiate itself in this uh in this very interesting
market we we we uh we're we're in at the
current moment.
Speaker 1 (56:56):
Hopefully one of those can help us to inflect inflect
the Gartner curve. Yeah, absolutely so Sean Christian, thank you
so much for joining us on the Vanguards of Healthcare
podcast today
Hi everyone, and welcome to the Vanguards of Healthcare podcast.
I'm Andrew Galler, one of the senior biotech analysts with
Bloomberg Intelligence. Bloomberg Intelligence is the corporate research arm of
Bloomberg that encompasses five hundred analysts worldwide. I'm joined today
by Sean McClain, founder and CEO of ABSI, and Christian Stegman,
Senior Vice president of Drug Creation Shan. Christian, thanks so
(00:40):
much for joining us today.
Speaker 2 (00:41):
Yeah, thanks for having us here. Andrew, thanks for having us.
Speaker 1 (00:44):
Maybe to kick it off, I'm saying it's a bit
relevant in the news. A few weeks ago, an article
from stat News came out claiming you and some of
your peers were mischaracterizing your work as Denovo antibody design
through jen Ai.
Speaker 2 (00:55):
How would you respond?
Speaker 3 (00:57):
Yeah, so, I guess I'll start off with the the
thing that I agree most with with the staff reporter. Uh,
two years ago, definitely was using too much hair jail,
So hopefully today the hair.
Speaker 2 (01:11):
Is looking all right.
Speaker 3 (01:13):
But in all, in all seriousness, let's let's just dive
into the facts. Two plus years ago, we came out
with a really exciting manuscript showing that we were really
the first company to put out work in antibody Danovo
design where we could design the h CDR three region,
the most variable region of the antibody, to bind to
(01:35):
a particular target of interest. Again, this was the first
time anybody had shown any work in antibody Danovo design.
And you know, two years later, we're at the point
now where we can actually use these models to go
after really hard challenging targets where there is no reference molecule.
(01:56):
And a perfect example of this, which actually was not
covered in the STAT article even though we had talked
to the reporter about this was the work that got
unveiled at our R and D day in December, and
this was work in collaboration with Caltech doctor Steve Mayo
(02:16):
and doctor Pamela Workman, and they had identified a region
in the HIV virus called the Coldera region. This is
a highly conserved region and today it's been very difficult
to drug. But the belief is if you can drug
this Coldaa region, you could actually potentially create a neutralizing
(02:37):
antibody to all the different clades or variants of HIV.
And we were able to indeed use our model, our
Denovo model, to design an antibody to this coldera region
completely from scratch. There was no reference molecule there. We
(02:57):
designed all six CDRs and UH, this UH I think
really highlights UH where we want to be with Denovo design,
being able to unlock a new novel biology by being
able to go after and drug these very hard and
challenging targets, just like we did with the HIV coldair region.
(03:20):
Actually have two other examples of being able to do this, UH,
our partnership with Omorole going after an ion channel and
then with with astrozeneca, a very challenging UH target as
as as well. UH. And So no matter what definition
of Denovo you use, UH, the work we did with
cal Tech UH meets any definition of Denovo that that
(03:46):
you could potentially have. And we're really proud of that work.
And so at the end of the day, UH, the
work speaks for itself. Put it out there, UH and
UH and and and it's really exciting and and at
the end of the day, where we're at with the
Denovo design of anti bodies fits really well into our
overall corporate strategy moving forward, which is to go after
(04:11):
targets that have known biology but have been difficult to
drug and this, you know, we believe is going to
ultimately give us an advantage being able to create you know,
first in class and best in class assets. Yeah.
Speaker 1 (04:23):
Absolutely, that's it makes a lot of sense, and that's helpful. So,
I mean, I think just in general, biologics are typically
viewed as a simpler construct than small molecules. Given you
don't have as much rug, you don't have any off
target interaction, you don't have to worry so much about
the metabolism of it.
Speaker 2 (04:37):
Do you think I have some mischaracterization?
Speaker 1 (04:39):
And then kind of what variable is do you think
you can tune in a biologic and how does it
impact the outcomes?
Speaker 4 (04:44):
We think biologics are actually not a simple construct. If
you look at, for example, sequence diversity, just the combinatorial
space for biologics, that's actually larger by most estimates than
the possible space for small molecules. Although it's very quick too.
You're going to be very quick to reach very high,
very large numbers very quickly. But if you look at
(05:06):
a monoclonal antibody, which is the you know, the workhorse
in biologics, you know, it's one hundred and sixty kilodoten protein.
It's a pretty complex protein. So the combinatorial space in
biologics is pretty large, and that's also where the strength
of generative AI comes in. So the ability to really
(05:27):
sample this sequence space close to exhaustion is really the
opportunity we have using generative AI.
Speaker 1 (05:35):
Absolutely, and then maybe I think apps that refers to
their biologics pipeline as smart biologic So what makes these
smart in your view?
Speaker 4 (05:44):
Yeah, absolutely, if you think about you know, the need
in biologics. To date, Essentially the large majority of biologics
have been generated using either mouse immunizations or display technologies
such as phage display. These technologies traditionally are always limited
by so called immunodominant epitopes in the case of imanizations,
(06:07):
or by whatever is generated using a display technique. So
the ability to design smart biologics is on the one hand,
the opportunity to actually consciously select an epitope that displays
the pharmacology of interest. And then on the other hand,
you can go further. You can engineer multi valency, you
(06:30):
can engineer conditional binding for example, pH dependent binding. You
can engineer certain multi specificities that are quite interesting. And finally,
obviously you can you can enhance the mode of action
by tuning potency and tuning affinities and then finally also
(06:51):
tuning functional activities.
Speaker 1 (06:54):
And you touched on the conscious selection of epitopes there.
So how do you prepare your approach to maybe some
of your competitors they want to do everything soup to
nuts fully automated in their AI models, because I believe
apps I basically uses drug hunters to decide their epitopes.
Maybe you can speak to that.
Speaker 3 (07:10):
Yeah, at the end of the day, like there is
still a human intervention that that's needed. Uh. And and
that's because we don't have the data to go full
end to end at at at.
Speaker 2 (07:21):
The current moment.
Speaker 3 (07:22):
If you look at just kind of the history of
of ABSI, we we started off as a as a
data first company. We figured out how to scale protein
protein interactions, you know, essentially the drug functionality, how the
antibody interacts with the target of interests. We're able to
scale that from thousands of interactions to to two millions
UH to ultimately use this data along with other structural
(07:45):
based data, to ultimately build out our Denovo model where
you can you know, design an antibody from from scratch,
but you have to have that that data to you know,
ultimately build these overall models. And we're just not at
the point right now where there's enough you know, functional
(08:05):
data out there to be able to have an AI
model be able to predict the epitope that's going to
give you the functionality.
Speaker 2 (08:14):
Now, ultimately, that's the direction.
Speaker 3 (08:15):
That ABSIT is wanting to go, is is instead of
a drug hunter or a structural biologist picking the target
of interest, we can actually have the model start to
to predict that and and and those are some of
the uh, you know capabilities we're building out right now.
Is is how do you you know, scale the overall functionality?
(08:37):
Uh so you can actually use the models to predict
the epitopes in the future.
Speaker 2 (08:42):
But right now we're we're using the AI models.
Speaker 3 (08:45):
To design the antibodies to the epiitopes that we think
are going to have the biggest impact on functionality. But
again we're we're not in this world right now. We're
design everything with a with a click of a button.
I think you know, in five plus years right now
we will get there. But there is that that human
intervention that is absolutely critical to the success of of
(09:08):
AI uh for for you know, for us and I
think others.
Speaker 2 (09:13):
In this in this field as well.
Speaker 1 (09:14):
And that's the data for the epiitope side. But let's
talk about the data for the antibody sequence side. I
think some of your competitors have talked about using public
data sets and the prominence of maps without mutations because
they're derived from naive B cells, so that reduces the
predictive ability for any sort of German mutation you introduce
into a sequence.
Speaker 2 (09:33):
How has apps I worked to overcome this.
Speaker 4 (09:35):
Yeah, you're totally right, Andrew. The scarcity of data generally
in the antibody space is the limitation because obviously you
need to train your models to really be able to
journalize across different target domains. And you know, beyond the
mutations you just mentioned, we think actually the bigger the
(09:56):
bigger issue is the lack of structural data in a
public domain. The public domain right now has around seven
hundred structures of antibodies with their cognate antigen So that's
clearly magnitude a way of what would be required to
have a fully generalizable model, that is that can be
(10:17):
meaningfully trained. But that said, there are ways to address
this obviously, you know, we have been very conscious in
generating and acquiring data sets in addition to the publicly
available data sets to enhance our models. And then we
also use additional technologies, for example, more likectodynamics to first
(10:39):
further diversify and validate our data sets. So I think
it's at the moment it's very much about being being
smart and very very conscious about what the data currently
entails and how to enhance it. And I think in
the future we'll see a lot more data becoming available.
But at outside we view data as finally as as
(11:01):
the final mode of being successful using generative AI, and
that's what we've building been building over the years. We've
been building a large set of affinity sequence data pairs
that ultimately allow us to really uh improve our model significantly.
Speaker 3 (11:21):
Yeah, and and just to kind of you know, hit
on that a little bit more, I mean, you know,
not only are we using your publicly available structure data,
we we have our obviously our sequence data where we're
generating you know, in house. And I think a direction
we're actually really excited about and in the future that
that Christian hit on is the the MD simulation and
actually using MD stimulation to generate synthetic data as as
(11:45):
as well to to enhance the overall generalizability of of
the models. And and so you can see that, like again,
data is so key to getting to the point of
generalizability within biology, and that's has been a huge focus
for us, whether that's experimental data or generating synthetic data
(12:06):
through MD simulation.
Speaker 1 (12:09):
A broad philosophical question almost do you think smaller bio
tech biostartups are kind of a disadvantage on the data
side versus these kind of large cap pharma companies that
have decades of development experience and even if all their
data is untagged, you would assume it's somewhere there in
the background.
Speaker 3 (12:26):
Yeah, I can hit on this and then you know,
Christian can you know, speak to his his large pharmat experience.
But I think what ends up happening in large pharma
is that, yes, they generate a ton.
Speaker 2 (12:36):
Of data, but it's very diverse data.
Speaker 3 (12:40):
It's it's you know, there's not you know, you know
a lot of you know, let's say, structural based data
on a particular problem you you want to solve, and
so you know, to solve a you know, a particular problem.
Like Denovo design, you have to have a lot of
a you know, specific kind of of of data. And
I would say that, you know, pharma again doesn't have
(13:03):
you know, a you know, in some cases they have
a lot of you know, specific data, but it's it's
more broad, uh, data across you know, many different types
of of data.
Speaker 2 (13:15):
And so there's only so much that.
Speaker 3 (13:17):
You can you can ultimately, uh, you know, leverage that
and and use it for And I think you're seeing
pharma do that and and in the areas that they
don't have the data and they're not getting the generalizability
on the models.
Speaker 2 (13:29):
You know.
Speaker 3 (13:29):
I think you're seeing them you know, partner with with
biotech companies, uh to to be able to kind of
build out those those capabilities. But uh, I mean, uh
a Christian please step in with the you know, your
experience at at at at buyer.
Speaker 4 (13:43):
Yeah, thanks Sean. Now, just to add it to what
Sean said, I spent more than a decade in big Farmer,
so I think I know what what how this looks
from the inside. I think you're not going to be
surprised to hear that there is, indeed a lot of
data available in big pharma and on paper that should
give them an advantage, but the reality is that usually
(14:06):
that data is not accessible, and it's not available in
a format that actually allows machine learning models to be trained.
So most of the big farmers actually going through a
massive digitization and transforming digital transformation process to make this
data available and actually make those available for machine learning models.
That said, I think the overall magnetude magnetitude problem that
(14:29):
I alluded to earlier remains right, because obviously the amount
of data each single farmer company has internally is still
a fraction of what's available publicly. So I think even
if they're successful with that transformation effort, the overall, the
overall amount of data is still is still limited. So
(14:51):
I think you know to your philosophical question earlier. I
think in the end, what we decided to do is
read to be very cognizant about what we want to
generate in terms of data that is really additive to
an genitive AI model, and then be very thoughtful about
how can we complement that with synthetic data, for example,
(15:12):
using molecular dynamics. We think that actually puts us in
a bit of an advantage because we are very quick
to it to rate and we're very thoughtful about how
to make the best out of these models. So I
think it's a it's an exciting race to be in.
I don't think we're at this advantage being a small
company at all.
Speaker 3 (15:32):
Yeah, And I think one thing in addition to being
a small company and having kind of the data platform
that we have is it's allowed us to create this
this lab in the loop process where we can rapidly
go from data generation in the wet lab to training
our models to then go in and validating those and
we can do this in a six week time period,
(15:52):
so we can you know, almost like a tech company,
rapidly iterate on the model architectures, you know, the designs,
the hyper parameters to really kind of figure out the
ultimately the best data sets to be pairing with the
best moldor architectures. And this has given us a huge advantage.
And I think that that's another reason, this lab in
the loop why we've in two years been able to
(16:13):
go from you know, designing you know, uh an h
cd R three of an antibody to designing now all
six CDRs to challenging targets. Because we've been able to
have this lab in the lab in the loop, and
I think that that is very differentiated and and you know,
being nimble and fast and having that type of technology,
(16:33):
I think is you know, something that is usually associated
with a you know, smaller, more nimble company versus a
large uh you know, big pharma.
Speaker 1 (16:43):
And you touched on this proprietary wet lab that you
have as part of your lab in a loop. So
I guess maybe do you want to discuss how you
think it's differentially and how it fits into this kind
of iterative learning process and providing reinforcement maybe much faster
than other AI enabled companies would.
Speaker 2 (17:00):
Yeah. Absolutely.
Speaker 3 (17:01):
So we're you know, we're able to ultimately, uh you know,
screen the these protein protein interactions. Uh, you know, in
a very uh you know, fast and effective manner essentially
how antibodies are are interacting with with targets of interest.
Speaker 2 (17:17):
Uh.
Speaker 3 (17:17):
And you know we can you know, we can scale
this up to to generate you know, millions of interactions
to you know, for for training. But another key piece
of this this technology is that you're able to actually.
Speaker 2 (17:29):
Use it for for validation.
Speaker 3 (17:31):
Uh, so you can actually see how accurate your your
your models are uh and and go and test, you know,
hundreds of thousands of uh discrete, UH specific sequences that
that are coming out of of your model. Again to
to see how accurate your your your models are. And
so we're using it both on on the training side,
(17:54):
but also importantly on.
Speaker 2 (17:56):
The overall validation side.
Speaker 1 (18:00):
Maybe just I know they're probably not statistic behind this,
but has that affect the actual outcomes and the I
guess closeness of the model to the actual outcome relative
to some companies that might just do purely in silica
reinforcement learning for their models.
Speaker 4 (18:15):
Yeah, look, I mean it's it's difficult to compare us
to others because you know, we don't know what what
hit rates others have. We've published our you know, in
some of our pre prints, we've published our hit rates,
and I think they are competitive and they are they
are in particular when it comes to targets for which
(18:35):
classical technologies have have failed, I think they are quite remarkable.
So again, coming back to the Caldera example that Jean
mentioned on the HIV Buyers, I think that's something that
has never been achieved before, and I think that's a
testament to the fact that you know, this technology allows
us to really address unmathematical need in a way that
(18:59):
has not been possible before. So for us, you know, AI, Gennifer,
it's not just a tool we want to use because
it's fun and it's you know, it's fancy, but it's
really the opportunity to deploy it in a way that
you know, addresses un mathematical needs.
Speaker 1 (19:17):
Okay, And then I think when we talk about small
molecule companies with AI, they talk about the chemical space
ten to the sixtieth, ten to the twenty third whatever
their estimate is, and how they want to explore the
unexplored chemical space. What do you think the analog is
for biologics? Is it just antibody sequences or maybe you
can take it from there.
Speaker 4 (19:36):
Yeah, sure, I mean the classical way a protein engineer
views the combinatorial space and an antibody is you essentially
take the complementarity determining regions of an antibody and you
essentially take the entire sequence space that's possible in these
six CDRs that a human antibody has, and then you
(19:58):
permute all of those.
Speaker 2 (19:59):
Right.
Speaker 4 (20:00):
Obviously, the human immune system has limitations it can it
typically does not permute all those in the computer we
can do that. Obviously, we need to pay attention that
we stay human humans is critical because because we don't
want to a generator molecule that is immunogenic, has toxicity liabilities,
has the propensity to form anti drug antibodies. But within
(20:22):
that boundary, and our models are obviously trained on human sequences.
Within these boundaries, the addressable sequences space is enormous and
I think that's the that's the opportunity. So depending on
how you look at it, you could you know, ballpark
the sum of all ammune acids across all CDRs is
(20:44):
probably anywhere between fifty five and seventy and then you
take that number to the power of twenty, which is
then the number of amino acids you have, So that
gives you roughly the combinatorial space you need to substract
that those sequences that are not human, right, that would
be sort of a back of the envelope calculation.
Speaker 3 (21:06):
Yeah, and I and I think like one of the
advantages to that that we have being in the you know,
in in the biologics space is that we I guess it's.
Speaker 2 (21:17):
A plus animus. You have you know, living organisms.
Speaker 3 (21:20):
That that have to actually you know, produce the protein
of interest versus a you know, medicinal chemist, and so,
you know, I think some of the issues on the
small molecule.
Speaker 2 (21:29):
Side that that that arise is that.
Speaker 3 (21:31):
You can you know design, you know, use generative AI
to design these really exciting you know small molecules, but
they just may not be you know, possible to to
actually make. And it's very time consuming to to have
a chemist go and synthesize you know, all of these
very you know, specific chemistries and hits that the model
generates versus you know, we can easily again using our
(21:54):
own wet lab technology easily you know, build out you
know one hundred thousand and uh you know sequences that
the model produced and and have you know our you know,
our platform manufacture those and and go and test and
and screen those again in kind of that six week
uh you know time period, and so we can kind
of make that that rapid uh iteration I think much
(22:17):
faster than the you know, small molecule folks can can do.
So I think there's a bit more flexibility, uh and
and ability to to iterate uh and and and and
go kind of through these you know lab and the
loop processes you know, much more with biologics than small molecules.
Speaker 1 (22:38):
And then I want to get into your individual assets,
but maybe just one high level question. When you think
about AI's potential to change the drug development paradigm, is
it time and cost and the pre clinical stages or
do we think that you can change the probability success
for drug development more broadly?
Speaker 3 (22:52):
Absolutely, Look, we're we're focused on solving uh, hard problems
that that that still exist. I think, yes, it's it's
great to do things is you know, faster and cheaper,
but that's not why we built out our r AI platform.
We we really built it out to solve the challenging
problems that still exist with within biology going after these
(23:13):
these undruggable UH targets. And and by being able to
go after those undruggable targets, you you have the ability
to now address uh, you know, diseases and indications that
you know previously drugged or or or solved.
Speaker 2 (23:28):
And so I think that this is why we we do.
Speaker 3 (23:31):
What what what we're doing, and and you know, ultimately
we want to utilize this to create you know, first
and class, best in class assets and and ultimately, yes,
increase overall probability of of of success and and uh,
you know open up the the you know, the different
diseases that that are possible to ultimately solve.
Speaker 1 (23:51):
Okay, great, So let's talk about a b S one
on one, your lead wholly owned asset t O one A.
So you're coming from behind against some large editors to Barley.
Where do you see the holes in the current product
profile from your rivals?
Speaker 4 (24:05):
Sure? So what we what we set out to address
with ABUS one on one when we started the project
eighteen months ago, as we looked at what's already out there,
there are three competitors. There were three competitive molecules, as
you mentioned, and all of them had liabilities. Some of
them has have liabilities in terms of feminogenicity, which means
(24:25):
a high propensity of a patient developing anti drug antibodies,
which may then lead to a lack of efficacy in
these of these drugs, over insufficient pharmacokinetics or requirement to
dose more frequently a requirement of so called induction dose,
(24:46):
which is intravenious. And then finally there's also a question
regarding efficacy and the ability to dose higher. So we
looked at all these parameters and we define a target
product profile for ABS one oh one that addresses all
of these shortcomings. So in essence, what we've been able
to achieve with ABS one one is a more potent,
(25:09):
high affinity molecule that has very good developability, which means
it can be formulated a turn a milligrams permil dad
and in fact we already achieved that, So even before
going into first in human we have already now achieved
a turn a milligram per formulation, which is usually what
what others do much later in the clinical development plan.
(25:32):
We've also addressed the umunugenicity issue. We have pre clinical
data that demonstrate that we have much less of such liabilities.
And then we also engineered half life, which means our
antibody is capable of achieving a much longer presence in
the plasma, potentially allowing every two to three months dosing.
Speaker 3 (25:54):
I will not too in the NHP studies, we are
also seeing the potential for you know, better biodistribution and
and so we believe with that we could also potentially
eliminate a loading doses as well, which you see with
some of the other competitor molecules.
Speaker 1 (26:11):
And it's one thing about your designs that a BS
one oh one has a dual affinity for both the
TL one A trimers and the monomers, whereas most of
your competitors focus more so on the trimer. So why
do you view monomer activity as important?
Speaker 4 (26:24):
Yeah? Absolutely, we get that question a lot, and it's
it's interesting, U. There's actually published literature out there that
that seems to be somewhat overlooked that clearly shows that
there are isoforms and fragments of tier one A that
remain monomeric through protioli protiolotic cleavage, and those can no
(26:45):
longer form traumeric structures. However, these short of fragments have
been shown to elicit subtle effects. In other words, they
can induce apoptosis and they can stimulate into fer comma release.
So we think it's reasonable to assume that in patients
with increased levels of trimeric to one day, there will
be also an increase in monomeric to one day through
(27:06):
these shorter isoforms. And we believe that having the ability
to bind the monomer as well as a trimer could
pretend potentially result in a treatment benefit.
Speaker 1 (27:21):
Then you touched on the pre clinical experiment that showed
low complex internalization, which should drive lower rates of antibody
anti drug antibodies and less imunogenicity. But as that panned
out in clinical data, we've seen the date for your competitors, we.
Speaker 4 (27:35):
Can only talk about what the competitors have disclosed. So
for the ROSI molecules, we've seen in a phase one
study that fifty six out of sixty eight patients have
developed anti drug antibodies, and that's the molecule where we
think we see also an increase in internalization. So we
think there is a pretty good correlation between that internalization
(27:59):
assay and in the end clinical ada rates. It's interesting
that's actually not only the case for for Tier one A. UH.
There's a more historical example of this effect out there,
which is on tinafalfa. There's a there's a molecule called
inflix mapp inflex the map for which this effects has
been described for the first time. So the inflex mapp
(28:22):
immune complexes are also quite.
Speaker 3 (28:27):
UH.
Speaker 4 (28:28):
They have been described to elicit antidoch antibody formation as well,
so there's precedence and we think the same effect you
see an inflexing map will apply to a Tier one A,
which is why we have selected an epitope that has
much less of this propensity to drive immune complex driveniminotenicity.
Speaker 3 (28:48):
Yeah, and kind of going back to to the AI
platform and uh, you know, we we utilize the AI
platform to design the antibody.
Speaker 2 (28:57):
To a particular epitope that we thought was going to.
Speaker 3 (29:00):
The immune privileged and and you know, ultimately wasn't going
to elicit uh, you know, a B cell.
Speaker 2 (29:07):
Response. Uh.
Speaker 3 (29:08):
You know, based on the clinical data you saw out there,
you know, MERKS eighty A rate was was was much
lower than than the roy VAN and if you looked
at just the T cell activation of of those, you know,
they they were very similar, which again led us to
believe that this was B cell mediated. And so we
designed an epitope you know, closer to the MRK epitope
(29:31):
because we we thought that that was an immune privileged
site that you know, ultimately would would elicit a much lower.
Speaker 2 (29:40):
B cell response.
Speaker 3 (29:42):
And again, so that's kind of a way that we
utilized our AI platform to engineer in uh, you know,
a specific attribute in this case, you know, potentially lower immutagenicity.
Speaker 1 (29:56):
They made another question about epitope choice. So t O
and A as aand has two receptors, DR three and
then d c R three or decoy receptor three. So
your competitors have talked about preserving binding of t O
one a to d c R three as basically a
sink of free t O one a. Do you think
that's important for ABS one O one and does it
do that?
Speaker 4 (30:15):
Yeah, very good question. So we have looked at d
R three and DCR three binding, and we have also
obviously tested all known antibodies out there, and we essentially
determined the racials of d c R three and DR
three binding. We think ABS one on one has demonstrated
(30:37):
the best ratio of DR three tow DCR three binding,
so the best selectivity profile at least in our hands,
and we think this could be a differentiating feature. On
the other hand, it's in a way arguable because the
DCR three concentrations inpatient in patients is actually very very low.
It's one thousand and four lower than the doses at
(31:00):
which the monoclonal antibodies are administered. So arguably, if there's
a sync effect, it's probably negligible. But if there's a
sink effect and it's meaningful, then ABS one oh one
has actually the best selectivity ratio.
Speaker 1 (31:13):
Then maybe maybe past the antibody design and thinking forward,
some of the competitors have adopted biomarker led approaches for
proof of concept trials, whether that single nucleotide polymorphisms or
soluble too one A levels. Does AP think that that
you're going to take a similar tact and how important
do you think identifying these potential super responders are?
Speaker 4 (31:32):
Yeah, obviously that's a super interesting field. We've watched it
very closely. You have probably seen the recent data published
around to the soakeybart in the Artemis you see study,
and in that study they showed no difference in outcomes.
In other words, an enrichment at least using the similar
(31:55):
nuclear tide polymorphis since they have selected in that study
did not result in any and in any treatment benefit.
So it's it's at this point I think, still a
valid idea, but it's actually very difficult to put it
in practice. The other piece that that is still you know,
out there for for the market to to to decide
(32:15):
is how how is the uptake going to be with physicians? Right, so,
to what degree will a physician based treatment decision on
such a test? I think an all comer antibody actually
has a big, big advantage there because if you are
able to demonstrate very good afflicacy and an all commer approach,
(32:38):
and I think it is one one is is very
well positioned to deliver such a target product profile. You
know that you can argue that you could those even
higher to to deliver a better efficacy. So, in other words,
I think the the all comer approach is probably going
to be the one that's going to be more attractive
(33:00):
and more successful in the market.
Speaker 2 (33:05):
Yeah.
Speaker 1 (33:05):
Absolutely, So we talked about aside AI platform for antibody
design and everything else a lot about the companies we've
talked about using AI in their clinical trial conduct and design.
Does Abside plan to start integrating that as well as
you move into the clinic.
Speaker 4 (33:20):
Yeah, Look, I mean this is a really exciting field. Certainly,
the entire value chain of research and development and the
pharmacy real industry is ripe for disruption by artificial intelligence.
We are exploring actively how we can use genitive AI
in such processes as well. I think it's fair to
(33:41):
say this is not our focused area right now. We
are laser focused on delivering biologics using genitive AI. But
we are certainly exploring this, and I do think we
will see a lot of innovation in these fields going forward. Yeah,
and I think it will span the entire value chain,
starting from patient selection all the way through regulatory submissions
(34:04):
and clinical trial protocols. So there will be innovations in
the field and we actively exploring those as well.
Speaker 2 (34:11):
Yeah.
Speaker 3 (34:12):
Look, as as Christian said, you know, we're we're not
pursuing this ourselves, but you know, these are you know,
areas that you know, we we see as as being
you know, you know, hugely transformational. And I think these
are areas definitely that you know, we see ourselves, you know,
partnering you know, with with leading AI clinical development you know,
companies in the future. But I think you know, given
(34:35):
the market you know, dynamics, and and the need to
you know focus uh, you know, we've decided to focus
in on on the Denovo design of antibodies. But again
I think, uh, you know, AI is transforming across the board,
and we're looking to partner uh with with folks that
are you know, you know, partner or that are utilizing
(34:55):
AI not only in in uh you know, clinical development,
but also you know, potentially target discovery and anywhere we
can integrate it to ultimately, you know, get the best
outcomes possible.
Speaker 1 (35:10):
Maybe can you just give a quick status update on
ABS one on one I woul understand you're still an
I in d enabling studies.
Speaker 2 (35:15):
When do you plan to enter the clinic?
Speaker 3 (35:17):
Yeah, so we're going to be entering that the clinic
here within the next couple of months, which is really exciting.
And we plan to have a Phase one interim readout
in the.
Speaker 2 (35:31):
Second half of this year, in.
Speaker 1 (35:34):
Which we expect from an interim readout given a Phase
one will probably be in healthy volunteers.
Speaker 4 (35:40):
Yeah, so what we will read out and this is
probably not surprising, it has been well described for Tier
one A. You can measure solubil Tier one A in
plasma and in healthy volunteers you can expect an elevation
of solubility one A when it's liberated from the tissue
bound by the antibody and then getting back into the
sexual relation. That's a very reliable and de risking biomarker.
(36:05):
It's a target engagement biomarker, and in particular for our
extended half life proposition, we hope to see a sustained
elevation of solubility one day in healthy volunteers after a
single DOLLS. We think that's going to be a significant
d risking of the program overall.
Speaker 3 (36:22):
Yeah, and you know, we we recently at JP Morgan
presented the target engagement data that we were able to
obtain from our non human brainmate studies and and that
was a you know, a really beautiful profile of of
you know, our our target engagement compared to the the
competitor molecules. And and that you know profile is a
(36:46):
similar profile that we're hoping to to see in uh
IN in our phase one interm readout.
Speaker 1 (36:53):
As Christian was mentioning, and maybe after you think on
the other side of this interim analysis for the phase
one and as you move in to prove a con
axcept an individual indications. IBD is obviously kind of the
classical TiO one A indication, but we see now that
some of these companies are starting to expand into other indications,
other inflammatory conditions. So how aggressive is appside planned to
(37:14):
be an indication to both in IBD and also outside IBD.
Speaker 4 (37:18):
Yeah, So TILL one A is a super exciting mechanism
because it has a lot of potential outside of IBD,
and we are absolutely looking at indications outside of IBD
as well. We have not this closed publicly what our
plans are specifically, but I can assure you that beyond IBD,
(37:39):
I think the market potential will be significant and hence
it would be improvedent not to explore it. So that's
what we're doing. In the end, we will go a
route that allows us to derisk the program as fast
as possible and then allow us to move into late
stage clinical trials in an indication where this more think
(38:00):
you can deliver the highest value.
Speaker 3 (38:02):
Yeah, and I think it is you know, one thing
you know, I will note I think it is important
to be able to to show potential, uh you know,
superiority to the clinical competitors, and so you know, definitely
at least you know, doing a a you know, proof
of mechanism uh in in UC uh to show you know,
how maybe S one on one compares to the clinical competitors.
Speaker 2 (38:26):
I think it is obviously really important.
Speaker 3 (38:29):
And then as Christians that we're looking to see, you know,
what are some other uh you know, indications we could
go after, because I think that there are some really
exciting opportunities to to to go after there.
Speaker 1 (38:41):
Maybe you can shift gears now and talk about ABS
two one, which you recently unveiled at your R and
D day in December. Maybe just walk us through the
genesis of this program because biologics for what are broadly
used as cosmetic indications haven't really been seen in the past.
Speaker 2 (38:57):
Yeah.
Speaker 3 (38:58):
Absolutely, you know, maybe just before we uh close out
one on one, there's one additional point that I that
I wanted to make on on one on one. Uh.
You know, if if you look at the duet you
know trial that janej is doing, it's a it's a
combo based you know trial in ib D.
Speaker 2 (39:14):
Uh. And and we really.
Speaker 3 (39:16):
Do you know, believe that you know, combo based or
or a bispecific approach is ultimately going to be the
way of the future.
Speaker 2 (39:24):
Uh.
Speaker 3 (39:24):
And you know, one of the things we're actually really
excited about that we've been developing is actually a Tier
one a bi specific and and this by specific is
not the the other arm is is not a known
target or at least a target that's been in discussion.
It's not an Aisle twenty three or an Alpha four
(39:46):
to seven.
Speaker 4 (39:47):
Uh.
Speaker 3 (39:47):
This is actually uh, you know, the other arm is
is actually a known I and I target that has
kind of been difficult to drug for for various reasons.
And you know, we think that this as as a
follow on to our our best in class or potential
best in class Tier one A and A body is
(40:09):
you know, potentially going to I think be a you know,
really exciting asset to follow.
Speaker 2 (40:15):
And and so.
Speaker 3 (40:16):
Anyways, wanted to cover that last little piece on one
on one before diving into two one.
Speaker 1 (40:21):
Yeah, I appreciate you throwing that in.
Speaker 4 (40:23):
Yeah, yeah, I can cover two one, So you're totally right, Andrew.
I mean, the aesthetics field is typically a very risk
averse field. That's why you have not seen a lot
of innovation there. Actually, the last novel treatment for androgenic
alopecia has been approved more than twenty years ago, So
this field has has a scarcity of innovation, which is
(40:46):
why we're very excited about it is two to one.
We think the market potential for energen alopecia is enormous
and it's actually growing. So ABES two to one really
has the potential to not only address hair loss, it's
actually a mechanism that potentially allows hair regrowth and also
(41:08):
address pigmentation problems. So based on the data we see
pretty clinically there could be an opportunity to actually get
back your natural hair color as well.
Speaker 3 (41:19):
Yeah, I mean if you talk to to dermatologists, you
talk to to patients that they're looking for a hair
regrowth uh product or medicine that can that can truly
you know, regrow their hair and and have a durable effect.
I think a lot of the medications that are out
(41:41):
there are are are hit or miss, and so if
you can you know, deliver to to you know, patients
a you know, a drug that can give them hair
regrowth as as well as the durability of of that,
I think that that, you know, that's something that that
patients i think are really yearning for. And as as
(42:05):
Christian said, it's it's a massive opportunity.
Speaker 2 (42:08):
I mean, eighteen to ninety million.
Speaker 3 (42:10):
Americans suffer from androgenetic alopecia and and you know, the
last you know medication that or the lestra that that
was reproved was you know, twenty five plus years ago.
Uh And and so they're they're they're you know, in
terms of the the competitive landscape, there's there's there's not
much out there.
Speaker 2 (42:29):
It's a big opportunity.
Speaker 3 (42:31):
Uh, and we see ABS two to one as as
being you know, hugely transformational here.
Speaker 1 (42:38):
And maybe just touch on your choice of target prolaction receptor.
So there is a relationship and a shared pathway between
the prolaction receptor and JACK stat pathways. So what level
of confidence do you have in prolactin receptor given JACK
and hibbers have had sidely mixed results in androgenic alopecia
compared to other forms of alopecia like alopecia areata.
Speaker 4 (42:59):
Yeah, yeah, yeah, absolutely. So first of all, it's I
think important that we distinguish these two different different diseases.
We have energenic alopecia, which is pattern hair loss and
that's really driven by androgen signaling, so it's a hormonal
path physiology. And then there's alopecia areata, which is an
audimmune disease that results in hair loss. And it's correct
(43:23):
that JACK inhibitors have shown mixed results in energenic alopecia
and they also have a problematic safety profile. So these
kindness inhibitors have black box warnings. Some of them have
been approved in alopecia areata, but clearly those are systemic
(43:46):
JACK inhibitors, so they will also inhibit the JACK signaling
in your immune cells which results in im you know, suppression,
and this obviously results in side effects such as urinary
tract infections, pulmonary infections, and other problems which actually resulted
in these Black Books warning. So we think protactin receptor
(44:07):
modulation is distinct from the JACK inhibitors. And the prolactin
receptor is first of all, largely specific at least addressable
with a biologic largely specific to the half vollicles. And
the second of all, it has a dual role. It's
not only acting through jacktool stigaling, but it's also acting
(44:29):
through STAT five signing. So there is clearly a differentiation
to JACK inhibitors.
Speaker 1 (44:38):
And you touched on the safety profile of JACK hibbers,
which is well known and well described. But I think
we have some other prolactin receptor. In a BIS, you
start to see increased in the serum prolactin due to
the inhibition of the negative feedback on petuitary levels. Has
absite considered a way to circumvent this and what risks
do you think could be carried with increasing prolactin levels.
Speaker 4 (44:58):
Yeah. So look, we have very we are very convinced
about the safety of targeting the productin receptor. There's human
genetic evidence that suggests that targeting the producting receptor is
very safe. There's a report out by by researcher described
a female individual who has a complete lack of lack
(45:19):
of productin receptor signaling and she's healthy UH. And note
that this is a lifelong human knockout. She's given birth
to two healthy children, and she has overall normal serum
electrolyte and hormone hormone levels, so no abnormalities, with the
exception that she was unable to lactate, which is not
(45:40):
surprising if you target prolactin. So that suggests that indeed,
in that individual there was an elevated productin UH level observed.
But that suggests that that elevation of prolacting in and
of itself is not a concern. So the other piece
you have to consider, we are are certainly not a
(46:02):
lifelong aiming to deliver lifelong treatment. And also you have
to you have to keep in mind that about logic
does not for the most part, past the blood brain barrier.
So the ability opportunity is that we are not to
(46:22):
a large extent not hitting the pituitary. We're not hitting
the central norvil system. The antibody will go into the
half vnicles and that's where the action will be.
Speaker 1 (46:33):
And I know you haven't shared too much about a
b S two one profile yet, but where do you
see the potential to improve on realistically? The one competitor
in the field right now the Buyer Hope product, So we.
Speaker 4 (46:44):
Are familiar with that molecule. We also know the liabilities
of that product. I think we try to address all
these liabilities with ABS two O one, and in essence,
what that allows us to do is really to deliver
an anti body that is highly manufactural and highly developable,
because in the end, we want to deliver a very
(47:06):
convenient product, which is a solution that is suitable for
subcutaneous administration, ideally a self administration in an auto injective
device that requires also solubility. We've shown very high solubility
of our antibody. We've also engineered half life extension in
that molecule, and through uh the half life extension, we
(47:30):
can also increase bioavailability, so increased organ exposure through an
increased transpseytosis effect. So if you take all of these
attributes together, I think ALS two to one is highly
differentiated to the existing antibody that's out there.
Speaker 3 (47:48):
Yeah, and the other issue too with that antibody is
that you know that the pad life on that is
pretty short. So obviously with what we have here, you
have an extent to patent life. And you know, given
that Hope is running this this trial in China, we
(48:10):
we do believe that we could be the the first
in the US market for approval, which we see as
another huge, huge benefit as as well.
Speaker 1 (48:24):
Is there anything else you want to mention about two
on one before we talk about HIV cluderra.
Speaker 3 (48:29):
Again, we see ABS two one as as our flagship asset.
You know, we we see in terms of overall market
size definitely greater than thirteen billion dollars a year. We've
been able to you know, to not only you know,
show you know, hair regrowth, but we we do believe
(48:49):
that this is going to have a durability. If you
look at you know, NHP studies that have looked at
anti prolactin antibodies, you know, after uh six months of treatment,
you see durability for four plus years. And and that's because
you're you're basically taking the follicle and putting it back
(49:10):
into the anagen or active growth state. And you know,
once you're in that state, based on your genetics, you
can stay there from from from anywhere from you know,
two to to six years. So that's you know why
we think we're going to get really strong durability.
Speaker 2 (49:25):
Uh.
Speaker 3 (49:26):
You know, you only have to dose for six months
and and you you have hair growth for the next
you know, two to six years. We see it as
a as a very safe uh you know product, uh.
Speaker 2 (49:35):
You know, based on the human genetics.
Speaker 3 (49:39):
Uh and uh and and and again it's it's a
it's a very large patient population that that we're serving here,
and so you know we're we're we're rushing to to
to get this into the clinic uh and and hoping
to you know, have some efficacy readouts on on that
next year.
Speaker 2 (49:57):
But again, we see this as a as.
Speaker 3 (49:58):
A really exciting asset and kind of our flagship outse
that's moving forward.
Speaker 1 (50:03):
So many HIV just leave it very open ended to
gauge your excitement around being ABS both as treatment, pre
pre exposure, prophylaxis, and maybe even curative in the sense
that they could potentially drive out viral reservoirs.
Speaker 4 (50:17):
Based on what we could show, we have something in
hand that has the potential to indeed be broadly neutralizing.
The region we targeted on the HIV virus is an
epitope known as the Caldera region, and this is a
highly conserved region. And if we are successful to actually
(50:40):
make this into an antibody that that works in a
broadly neutralizing way. We are running these experiments as we speak.
It could it could potentially be a very attractive broadly
neutralizing asset for for PA patients with HIV. So it's
something we are actively working on. We've we've an excellent
(51:02):
collaboration with our collaboratives at Caltech, as Sean mentioned, Steve
Mayo and Pamela Bjorkman, and it's it's a great privileged
and opportunity to be working on this with them.
Speaker 1 (51:12):
And then maybe just in our closing minutes, let's talk
about a few broad AI questions. One, where do you
think we're currently out on the Gardner hype cycle with
regards to AI and drug development?
Speaker 2 (51:24):
Uh?
Speaker 3 (51:25):
Yeah, I definitely think we're We're coming down from from
the overall hype.
Speaker 2 (51:31):
I think, uh, you know, we're we're I think getting.
Speaker 3 (51:34):
Close, not close to the trough, but maybe uh, you know,
not quite there. But I think we're we're probably going
to get to to the trough, uh, you know, probably
in the next one one to two years. And then
I think we're we're going to you know, start to
you know, come come.
Speaker 2 (51:47):
Out of that.
Speaker 3 (51:48):
But but at the end of the day, like this
is not you know, a fad that's gonna you know
ultimately you know, fizzle out. We we are actually seeing
how ai IS is transforming this this industry, not only
in what we're doing, you know, like going after you know,
the coldera region in HIV, going after these hard challenging
(52:12):
or un drug ble targets, but you know, we're we're
we're seeing big wins on you know, you know, clinical development,
on on.
Speaker 2 (52:17):
The target discovery side. I mean just just across.
Speaker 3 (52:20):
Uh the board of ai IS is is here to
to stay and uh and and I think it's going
to have a profound impact on not only that the
the you know, the R and D costs and time,
but ultimately probability of of of success. And it's ultimately
going to take uh you know time to you know
(52:41):
see uh you know these stats you know pan out
just due to the regulatory process. But you know, even
look at you know what we've been able to do
on on you know, the the speed and and and cost.
You know, we we were able to you know, get
to a drug candidate in ABS one on one in
in in fourteen months, uh, you know, and be able
(53:01):
to get in into the clinic for for roughly you know,
thirteen to fifteen million dollars. And you know that could
normally cost you know, a normal biotech or large pharma
you know, fifty to one hundred million dollars. And so
we're already seeing big savings there. We're able to see
you know, us go after undruggable targets and and so
I but you can see, I mean there's like the
(53:24):
skeptics that that are that are coming out and and
so again I think we're we're you know, we're coming
down from the from the overall hype as as seen
in the in the in the stat article. But at
the end of the day, AI is here to stay.
Speaker 1 (53:36):
And what are you most excited about what's coming next?
Both for the abside platform and the field more generally.
Speaker 4 (53:41):
What I'm really excited about is the opportunity to really
deploy AI in a way that allows us to deliver
assets that have been elusive, right so, really working on
previously undragable targets. I think this is the This is
the frontier we are tackling right now, and then the
next thing after that will probably be a way to
(54:03):
view the target space in a more combinatorial way. So,
for example, if you look at one of the most
recent successes in the biologic space was an antibody called
the mechism ap, it's actually targeting into Luke in seventeen
A and into Luke in seventeen a f so targeting
two iso forms of an interluken and they could show
that this antibody is superior to what's already out there,
(54:25):
So dual inhibition in that case made sense. Now you
can ask the question why target only two or why
target these two? And I think the opportunity to actually
view the target space in a way beyond one seeing
a target is super exciting, and biologics using generative II,
there will be biologics that can do this, and this
(54:49):
is I think the next frontier that we'll see.
Speaker 3 (54:51):
After that and you know, we live in a really
exciting you know time time right now.
Speaker 2 (54:56):
I mean you look at you know what's going on
in an AI.
Speaker 3 (54:59):
You've had the the deep zeek uh moment and UH
and and you know we've we've kind of experienced in
biotech a little bit of kind of this you know,
deep seek moment or ourselves. I mean over this past year,
you you look at, you know, all all the acquisitions
and licensing deals that are you know, happening from from
from China. They're they're not licensing assets from uh, you
(55:21):
know US, you know biotech companies. You know, I think
it was over thirty thirty percent plus UH you know
assets over the over the past year, I've been you know,
either you know licensed UH or or bought from from
Chinese companies. And and that you know really kind of
you know bigs. A question like how you know, as
as US you know biotech companies, how do we continue
(55:43):
to you know, differentiate UH and and you know create
you know, novel breakthrough UH assets and and I think
it's it's leveraging AI. It's not to do it you know,
faster and cheaper and to compete with with with China.
It's it's actually to do something completely novel, uh being
able to again go after you know, uh, you know,
(56:05):
hard to drug targets like the cold Are region or
even like the by specific that we're we're talking about
where it's like, you know, you have you know, agonism
at higher concentrations when you want antagonism, how can you
you know, eliminate that agonism and and focus you know,
solely on the antagonism uh and and create you know,
these differentiated profiles like you know, pH dependency or multi
(56:26):
valency to ultimately increase overall probability of of of success.
And and that's really again where we're focused in on.
It's not better and faster, it's it's it's how do
we utilize uh ai to to to solve these these challenging,
you know, problems. And that's uh, I think, uh, you know,
the future for for ab sign, I think how uh
(56:46):
you know, we're going to see the apps you know,
AI differentiate itself in this uh in this very interesting
market we we we uh we're we're in at the
current moment.
Speaker 1 (56:56):
Hopefully one of those can help us to inflect inflect
the Gartner curve. Yeah, absolutely so Sean Christian, thank you
so much for joining us on the Vanguards of Healthcare
podcast today
Host
Jonathan Palmer
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!