Why mAbs And Why Now? With Absci

Intro:
Welcome to TD Cowen Insights: a space that brings leading thinkers together to share insights and ideas shaping the world around us. Join us, as we converse with the top minds who are influencing our global sectors.
Brendan Smith:
Welcome back to another episode of Machine Medicine, AI in healthcare: TD Cowen's podcast series, where we bring you the latest and most important takeaways from the state of AI in the healthcare sector today. I'm your host, and TD Cowen Healthcare Analyst, Brendan Smith. Today, I'm joined by Absci's Founder and Chief Executive Officer, Sean McLean. Sean, it's great to have you. Welcome.
Sean McClain:
Thanks, Brendan, for having me.
Brendan Smith:
Yeah, so for any listeners that are new to our series, Machine Medicine aims to break down the use of artificial intelligence in healthcare into bite-size digestible points, one episode at a time, highlight the biggest misconceptions, and then recontextualize each piece back into the bigger picture.
Today, Sean and I are looking to dive into how and why monoclonal antibodies, i.e. biologics, are such a big focus for AI models and drug developers today. Also, what FDA's recent stance on AI antibodies could mean, and how Absci's platform and approach to drug development is really looking to meet the moment.
Sean, let's just dive right in. First, maybe let's start with a quick definition of what a monoclonal antibody actually is, but also, why they've been so wildly successful as commercialized therapies, and what is it about mAbs as a class that make them such a prime target for AI models and platforms?
Sean McClain:
Yeah, absolutely. An antibody is a protein made up of amino acids, and you have to have living organisms that make these proteins. Our interest in being able to leverage AI for antibody design is the complexity. If you look at an antibody sequence, there's more sequence variants in an antibody than there are atoms in the universe.
Being able to go from this paradigm of searching for a needle in the haystack to actually being able to create the needle, in our case, a biologic, we see that as transformational, and being able to now hone in on all the attributes you want, and being able to hit the particular epitope of interest that gives you the biology that you want, being able to look at the developability and manufacturability. One of the reasons antibodies have been so successful is they're essentially targeted missiles.
You can hit a particular cell that you want or a target. Genentech developed some of the first antibodies that hit the market. One of them in particular was Trastuzumab that went after Her2. It's quite funny to actually look at some of the rhetoric that was used back then and what folks thought of antibodies. A lot of folks thought that antibodies weren't going to be successful. There's a lot of skeptics out there.
I would say it's similar to the skeptics on AI. We truly believe that just like antibodies were successful, leveraging AI to design antibodies to un-druggable targets, and really be able to leverage it to address unmet medical needs, we believe that that is the future.
Brendan Smith:
This raises a lot of good questions. I guess compared to maybe some other modalities, when you're building AI models for monoclonal antibodies, are there important considerations for the different types of mAbs, and maybe different tissues or organ systems that you might be targeting?
Sean McClain:
Yeah, one of the beautiful parts about using generative design along with antibodies is that it's really target agnostic. We can go after any target or particular indication that has big unmet medical need. The input to the model itself is the structure of the target. We specify where we want the antibody to bind the epitope, and then the model then designs the CDRs, or essentially the fingers on the antibody that bind to the target of interest. This is what really makes this platform, again, a really exciting platform, because it is target and indication agnostic.
Brendan Smith:
I think, look, when we're diving into some of these different applications and different modalities that these models can be used on, I think there's inherently a lot that is less known, definitely underappreciated, misunderstood, whatever kind of adjective you want to ascribe to it. Maybe from where you stand, what do you think the investment community in particular maybe underappreciates or misunderstands altogether about what has already been accomplished to date, with some of these biologics-focused AI models, maybe relative to some other healthcare AI approaches?
Sean McClain:
First, speed and cost are a big focus. How quickly can you get these drugs into the clinic, and how can you reduce the overall cost? I think the investment community, I think, fundamentally understands how AI can really drive efficiencies on those two parameters. I think the area that the investment community maybe underappreciates is how generative design can actually unlock new novel biology. What do I mean by this?
Being able to go after known targets that have been difficult to drug with traditional approaches, such as ion channels. It's been very difficult to actually develop antibodies that can block ion channels. We are actually able to show in our partnership with Almirall that we could actually develop a antibody to a known target that's been known for 30 plus years, and for the first time ever, have an antibody that was designed with our generative AI model that could block that particular ion channel.
This really is exciting. Now, you have a first-in-class asset, and it's de-risked from a biology standpoint, because we've studied this biology for quite some time. You're not taking as much biological risk, but you have the opportunity to have this first-in-class asset, where you're able to address a big unmet medical need. We've been able to show as well, we can go after other challenging epitopes, such as the HIV caldera region, which potentially will lead to a neutralizing antibody across all different clades of HIV strains.
Again, being able to leverage generative AI to unlock new, novel biology is something that we're really excited about, and what we see as being differentiating.
Brendan Smith:
Yeah, I think this really gets at the forward-thinking aspect of this too. It's not just refining what's already been done. It's not necessarily just doing things in a slightly maybe more efficient or with a fine-toothed comb. It's actually able to add this predictive quality of things that just haven't actually been explored yet. I think this also ties into this idea of, okay, so relative to what has been done, where is this going?
Obviously, biologics have been around, this class of drugs is not particularly brand new, but when we look back at how these compounds have been developed and manufactured historically, what do you think are some of the most important updates and differences that using AI platform is enabling, maybe beyond just the brand new approaches here? Is there anything that you would particularly flag in that respect?
Sean McClain:
Yeah, one of the areas that comes to mind, especially when you talk about manufacturability, is the iterative process that drug discovery used to have to go through. You end up finding a binder that gives you the particular biology that you want, but then you find out that it doesn't have the manufacturability or developability. You do some slight tweaks, but that may change the efficacy that you want. Ultimately, you're taking suboptimal hits into the clinic.
What AI allows you to do is do this multi-parametric modeling, where you can look and hone in on various different parameters all at the same time. You can have the model predict your binding, your epitope, essentially the efficacy, as well as the developability and manufacturability all in a single step. That's allowing you to get the best, highest quality molecule into the clinic in a very short amount of time.
I think that that's one area that AI is really helping expedite and get the highest quality drugs into the clinic.
Brendan Smith:
Yeah, look, it came out to see the Absci plant just before the holiday season, I think it was past December too, and it kind of really helped put an actual visual to some of this too, right? I think some of us on the outside often struggle to really understand some of the logistics of how these platforms evolve over time, and really what it means on the ground day-to-day.
I know you've touched on this a little bit here, but I guess maybe particularly with platforms like yours at Absci, where did you think the biggest value add in the drug development process is? I guess when I say that, it's kind of very boots on the ground, is it speed, is it cost, is it efficiency, is it volume? All of the above? I guess for those of us not involved in this day-to-day, where is the biggest value add for all of it?
Sean McClain:
Yeah, absolutely. I think all the points that you mentioned are really important. If you look at our TL1A asset that was developed with our generative design platform, we were able to, from target selection to asset in the clinic, we were a little over 24 months. That normally takes five and a half years. Shaving off the time to the clinic by half is a huge accomplishment. Then additionally, we're able to invest $15 million into getting it into the clinic.
Normally, that costs large pharma 50 to $100 million just to get a drug into the clinic. We're dramatically shaving the cost as well. I think all of these things are important, but I want to double down on what I talked about earlier, which is we see the biggest value in being able to go after these un-druggable targets and create new, novel biology. We see that as really important in being able to compete in the current landscape that's emerged.
You see China, China is really showing that they can have a great fast follower approach, where once a new target gets disclosed, you have 10 Chinese companies developing high quality molecules that you're seeing pharma go and acquire. How do we leverage AI to compete in this new evolving dynamic that we have in biotech?
Our strategy is really being able to go after these hard targets that our peers can struggle to drug, as well as Chinese companies, and really be able to create these first in class assets against known biology. Our earlier stage pipeline that we're focused in on is executing on that exact strategy.
Brendan Smith:
Yeah, I think this really kind of lends itself to a couple of different questions. Depending on how you want to use some of these AI technologies and some of these platforms, the idea, presumably, that you would kind of build better versions of previously developed mAbs, versus, I'll use some of these GenAI capabilities to more or less create or help find brand new targets for mAbs. Presumably, there's some differences in building those models and different considerations for some of that.
I guess my question then would be, if a company is focused exclusively on maybe creating better versions of existing therapies, does that require a different level or modeling approach from AI versus maybe those that are maybe doing that also, but also really focused on leveraging the technology to discover and create new targets altogether?
Sean McClain:
Yeah, absolutely. I think that there's three different components that you talked about. There is the target discovery, there's discovering drugs from scratch or de novo, and then there's taking known hits or leads, and being able to optimize them. On the lead optimization, I think that that is an area that a lot of folks had focused in on, and is something that's getting adopted internally across large pharma, as well as biotech.
That ability to improve a molecule is a lot easier than actually developing the molecule from scratch. The models that we use on the de novo side are different, and then the data input is different as well. It's a lot harder of a challenge to give a model a target, and have it predict a binder that can bind to the particular epitope of interest. That's really where we're squarely focused. We have models that can do both de novo design as well as lead optimization.
In terms of the difficulty, is really on that de novo design aspect. Then what I would say is even more challenging, and where the field is evolving, and I think we're going to continue to see breakthroughs, is in the target discovery, being able to leverage AI to predict new targets to go after. It's really early on, and I think we're still needing data to help us predict that translatability. Is that target actually going to translate and have the biology that you want, and be able to address the unmet medical need you're going after?
That's a lot more of a complex problem, because you're looking at a broader biology problem versus a design problem with de novo design and lead optimization. This is the future of where things are headed is being able to predict targets. We're not effectively there yet on using AI in that manner, but I think that there's a lot of exciting opportunities there.
Brendan Smith:
Yeah, and I think speaking to the future, I think I would be remiss if I didn't bring up FDA. Obviously, it feels like almost every other day now, we're getting some news out of FDA when it comes to AI and the implementation of it, and just the agency of 2025's renewed outlook and being a little bit more amenable to integrating some of these capabilities. We know that FDA is looking to phase out animal testing requirements for monoclonal antibodies specifically.
They called that out a couple months ago now in the press release. Might also include some other undisclosed drug modalities in lieu of what they're now calling "new approach methodologies," or NAMs. Let me first ask you, I guess number one, how transformative do you think this new policy, new guidance out of the agency could be for drug developers? Then maybe on top of that, what does that actually mean for drug discovery players like Absci?
Sean McClain:
Yeah. We were really encouraged to see the administration focus on AI within the regulatory process, and how we can use AI to help streamline the overall drug development process. We want to continue to see that, and the current guidance that is out there, it is aspirational in some regard. Being able to eliminate animal models and be able to predict the efficacy and the translatability of drugs, we're just not there yet.
We're going to continue to need to generate new sources of data to help with that translatability to really predict that efficacy. When it comes to safety, manufacturability, developability, I think in a lot of regards, we are there or we're getting close to being there. I even look at the models that we've built, like our naturalist model, being able to predict manufacturability, and indevelopability, and immunogenicity, we've been able to leverage these models to show what sequences are likely to provide liabilities in those various different aspects, and that can be utilized right away.
Again, the efficacy piece, I think we're a few years out. We need new data sources and new model architectures to get us there, but all in all, it's extremely encouraging to see the direction that this administration is taking thing.
Brendan Smith:
Yeah, I think more than anything, as is often the case at the federal level, but frankly, even within scientific development, the years of progress often grind fairly slowly. To your point, we're kind of moving in this direction. I think a lot of conversations we have, people are expecting maybe within five years or so, we'll start to see kind of a hybrid approach, maybe less animal testing, although not zero.
I worked in an animal lab for upwards of 10 years at some point. It's not the most fun work. I think you'll be remiss to find very many people who want more of it, but all these kinds of things take time to find this really hybrid balance over the near term until we're in a place to be able to deliver on all of that. All right. Look, we've covered a lot of great ground today. It's a conversation I'm sure we're going to continue to have for the foreseeable future.
Before I let you go, one thing I do like to ask people at this point is if everything we've discussed today goes super far over someone's head, but they've made it with us this far, what's one point you would really want everyone listening in to remember and take away from our conversation today?
Sean McClain:
Yeah, absolutely. I will say that AI is here to stay. We're in the early innings. We are seeing, in particular on the design side of things, that we can start to leverage AI to tackle hard, challenging problems that still exist, and being able to address unmet medical needs, and really create first in class exciting novel biology for patients.
We're going to continue to see AI transform this industry being able to ultimately, I believe, in the future, be able to really come to a point where we have personalized medicine, where you're able to, for a given patient, be able to predict the target that will address their particular disease, and then use models like our generative design model to then design an antibody for that particular target of interest, and have it all tailored to that particular patient.
We're not there yet, but we're getting early wins on the board. I see AI being one of the most transformative technologies ever to hit medicine, and drug discovery, and development, and it's going to be really exciting to see what happens over the next five to 10 years.
Brendan Smith:
Yeah, I think actually, on an earlier episode of this podcast, we had [inaudible 00:21:06] use this expression, the green shoots of success. I think that's a great way to kind of think about where we're at right now. Obviously, we're starting to see things really coming through, and it's very encouraging if nothing else.
I think with that, I want to just thank you for hopping on, and talking us through all these different AI use cases and biologics. I'm sure we'll have a lot more to talk about very, very soon. Always good to talk to you, Sean. Good seeing you.
Sean McClain:
Yeah, absolutely. Thank you so much.
Intro:
Thanks for joining us. Stay tuned for the next episode of TD Cowen Insights.