The Future of AI and Drug Development: Insights from Absci’s Sean McClain and Andreas Busch

Welcome to FYI, the four-year innovation podcast. This show offers an intellectual discussion on technologically enabled disruption, because investing in innovation starts with understanding it. To learn more, visit arc-invest.com. Arc Invest is a registered investment advisor focused on investing in disruptive innovation. This podcast is for informational purposes only and should not be relied upon as a basis for investment decisions. It does not constitute either explicitly or implicitly any provision of services or products by arc. All statements may regarding companies or securities are strictly beliefs and points of view held by arc or podcast guests and are not endorsements or recommendations by arc to buy, sell or hold any security. Clients of arc investment management may maintain positions in the securities discussed in this podcast. Hello and welcome to FYI, the four-year innovation podcast. My name is Brett Winton. I am Arc Invest Chief Futurist and the host of FYI. With me today, I have the man, the legend, Charlie Roberts, Charlie is our chief investment strategist and a multi-time founder in the multi-omics space. Charlie, welcome to the podcast. How are you? Thank you, Brett. Yeah. Very well. Thank you. Yeah, I'm excited to talk about and show listen to some of the amazing insights from this great biotech team that we're talking to you today. Yeah. So today, we are, we spoke to ABSI, which is a company that is really driving for drug discovery with artificial intelligence. Charlie, why don't you share who we spoke with? We spoke with Andreas Bush, who's actually the chief innovation officer. And you know, I think one of the big failings we see sometimes in the machine learning four bio spaces is what you could think of as machine learning hubris, where maybe the whole mission and everything is being led only by machine learning and biotechnology, but not enough by bio insights. And so it's really good and reassuring to see people of Andreas Bush's caliber in the drug discovery, drug development world. So he's had major R&D roles. Now ABSI, as chief innovation officer in leading a lot of the R&D, but historically in many of the big farmer with very, very high execution, high quality. So he's been in Sonofi and Bayer and Chaya. His leadership has actually resulted in over 10 commercial drugs, which is a really high hit rate or just overall performance starting from bench right through to the FDO approval. And with several others in late stage clinical development. So he's a real drug discovery, drug development, firepower being brought to the machine learning four bio space, you know, quite rarely. And he originally studied pharmacology back in Germany in Frankfurt. And then, you know, we were also joined by someone, you know, very young founder leader Sean McLean, who I believe is still not only in his mid 30s and has already achieved a huge, huge amount. So he's the founder, CEO, director of ABSI. And he's been the CEO since way back in August 2011 and a member of the board of directors of the foundation. He has studied in University of Arizona, has a degree in molecular cell biology and originally co-founded the company. And, you know, I think we talk about it a little bit through the podcast, but one of the things I think it's very impressive about Sean is he's one of those very few founder innovators who will get out in front and just champion his own science and, you know, has the knowledge and the way we're all to do it. I met him originally a few times at different conferences, but one memorable time he was standing in front of his own poster talking about the science of ABSI to essentially a group of PhD students who are quizzing him sideways on the technology. And you know, that was really, really rare in someone who's leading and that's that company. So that was exciting to see. So the reason ABSI is interesting to us and generally the idea of AI Drug Discovery is, you know, as they'll describe, you can potentially compress the time and the amount of money that needs to be spent to get a drug from kind of like this is a target we're going after to this is something that's ready to go into the clinic and even go up front of that using artificial intelligence to actually determine which target you should go after with which or which part of the protein complex you should go after to address a specific disease. Charlie, do you think that this is like we have enough capability now for that sort of thing to work? Yeah, look, I think we are seeing a really historic change in the overall ROI and the efficiency and the ability to deliver insights and turn them into therapies than we've ever seen. You know, we're really seeing time compression, we're seeing massive increases in the amount of progress per unit of cost and time and just a higher hit rate overall and ABSI is really one of the companies spearheading that direction. And you know, even we're not alone in saying this though I think it's still rare and not widely priced in or believed by the market, but you know, even if you look at folks like Delight this year have just finally published saying actually, look, Pharma R&D efficiency is finally turning and this is even within Big Pharma, we believe that Bio Pharma R&D, you know, with the nimbleness that comes there is even more efficient and you know, quite a lot of data that that's the case. And ABSI is just really one of the examples where you can very tangibly stick numbers on how much better it is than it was before. And you don't need to tip the scales much in favor of efficiency to really have a massive leveraged outcome in the number of drugs approved per billion dollars that we're having. So it's very, very exciting. Yeah, and so then that translates into basically dollar in leads to more dollars coming back from an investor perspective. And potentially what I find exciting about the space is as you know, they have kind of wet lab in the loop. So as they are seeing success that should feed back into improving the models they use to then make their drug discovery even better. And so I have this working theory that like the ways in which right now drug manufacturers are big is because they have very strong distribution. This isn't a theory. This is like demonstrate like large cap pharma is basically a giant sales force attached to an M&A acquisitions team that then provides them with molecules to sell. And so but they got big because you need the scale to do the sales, right? And actually this could be a different vector that causes kind of like this companies to become big in the innovation side of drug discovery rather than that being a totally fragmented space where everybody has like IP rights over little like particular technologies or particular end points, right? Right. And the bringing of people with the firepower of drug discovery drug development expertise of someone like Andreas who's on the podcast also brings to nimble machine learning AI enabled biotech the ability to actually prosecute those products right through. And of course some of them will get partnered and it's been announced that I was already partnered with folks like AstraZeneca and in fact the very senior ex AstraZeneca executive many pangoluses on the board and is also on the board of biogen now that just got announced. And so but there's this with this compression with the ability to actually prosecute some of those programs right through and earn them own them. You know we believe that companies like outside really have the potential to become the scaled biotech platform companies of tomorrow like the regenerons and the genetics but in much faster of a time because of that compression and the efficiency gains and while some of these you know programs will be partnered many will be owned right through and if you look at the you know the sort of the famous numbers about the 2.6 billion or whatever it takes to fit a drug to the right through to approval that fact is in a lot of failure any individual drug can obviously be developed for much much much cheaper than that and we believe that that's coming down by maybe orders of magnitude eventually even from there and you know one of the things you mentioned the biology that's interesting to highlight is the biology platform so I'm so almost as these two tracks where a lot is about drugging known biology known targets much much better and there's a lot of headroom there that's a huge time and a huge opportunity there but but outside also has an angle on addressing maybe the biggest I may need if you look at the coming decades of cures which is around biology and they've got a very really very innovative and it appears highly so highly effective way of backing into those biological discoveries from the from using the human so studying the human in the way that the humans immune system has has built antibodies against the problem which is just a very nice innovative there's a couple of other companies doing something somewhat similar sort of you know enough of a validated space with a few folks going after but it seems that outside has got a really differentiated angle on that and I think we touch on a little bit in the discussion but the use of AI to model sequences both to analyze them and in this case sequences as in proteins and otheromics both in analyzing them and then predicting what would be a you know in a generous of sense that is an extremely good and underused use of AI and you know we see it in language elsewhere but it's it's just really entering the phase at which a few players are doing this in biology and and and outside one of the leaders and you know there's some other very interesting companies so Alex Reeves X Facebook AI has got a private company who's doing different but but again part of the same then philosophy and and also Jacob Uscarite X one of the transformer offers actually one of the transformer offers also has a very interesting company called Inceptive which is also using sequence AI models in that in his case over the use of over RNA but also taking that that philosophical view that these are just language and their sequences and in that sequences knowledge and you can learn that knowledge by using the sequence models much as a GPT does over language and you know we think that's a very exciting direction definitely watch this space and you know we think we'll see a lot of other discoveries in that in that direction yeah and you know of all of the areas that we study I would say that biology is the most data rich for so there's a reason to believe that having this magical new tool that can interpret and interpolate vast sums of data is going to have like a very impactful practical application in the biological space it's in some we've had an abundance of data the healthcare sector has more data than any other sector and but they just haven't done what to do with it or how to use it and so kind of the promise here is oh now we have a tool that we can plug into these data sets to then at least on the absi or the drug side to actually you know be smarter about developing drugs towards the thing that we're trying to achieve for the patient so we hope you enjoy the conversation look out for the the anecdote late in it where we have to end the mouse monopoly on antibody design that's what that's you know it's just like a little needle in the haystack out there for you to listen to Charlie thanks for thanks for joining us and enjoy the interview thank you Brett today we have a great conversation with me today I'm happy to have Sean McLean Sean you're the founder CEO of absi you found at the company in 2011 and and really absi is it the the the forefront of linking AI to drug discovery and development particularly in the biologic space and you've absi is positioned to really lead in that area and part of being a great leader is you've brought on great talent so Andreas Bush if I'm pronouncing your last name correctly I think I am has is is from a traditional farmer background you were 13 years at buyer as you know global head of drug discovery brought a bunch of products to market and chief scientific officer and head of R&D at Shire so Sean you've really taken somebody from the traditional space into the I'd say the next generation of drug discovery welcome and thank you for coming on yeah thank you so much for having us and I think we're going to learn throughout this presentation we're definitely very multilingual team it takes takes a whole army and village of folks that know a lot of different skill sets to truly make tech bio and AI drug discovery work Sean maybe we can start out with you could you lay out why like one why AI should be used to kind of drive drug discovery and why now like you've found at the company in 2011 but clearly they're we're in a different day in age than we were then how is absi positioned and how do you think the entire industry is going to change as AI changes the way we do drugs yeah absolutely and I think really where absi is focused is using AI to solve the design problem so you can you know scientists can now test hypotheses that they haven't been able to test before and and you know what what let's dive into that but let's look at you know these undrugable targets like gpcrs or or ion channels the these you know targets that have known biology but existing ways of of discovering antibodies or small molecules have been have been difficult in order to ultimately drug these these targets but what if you could actually take AI and design those those drug candidates or those antibodies to these undrugable targets starting to unlock new novel biology and and that's really where we're focused you know both on our internal pipeline but also with with partners you know partners like AstraZeneca you know Omarol and Merck these are you know you know pharma partners that have you know struggled with the same thing we're struggling with and you know I think these partnerships again are allowing us to leverage domain expertise that pharma has in therapeutic areas while leveraging our AI platform to to unlock the these novel biology and novel drugs that we can ultimately take into the clinic and you know and kind of you know stepping back a little bit you know talking about absi's you know history we were not an AI first company we built a synthetic biology platform that allowed us to scale protein interactions essentially how antibody based drugs interact with a target of interest and so to make an antibody efficacious or actually work you need to hit the right location on the target and and the right affinity and and and essentially this platform that we discovered a lot of us to so you know traditionally go from screening thousands of drug candidates in a given week to being able to screen millions and you know that was right around the time you know 2018 when transformers were we're taking off with with who of these first you know gen AI models and it was like this idea of like wow if we could take this this data of protein protein interactions with these generative AI models we could really go from this paradigm of searching for a needle in the the haystack to actually creating the needle in our case a a biologic or an antibody based drug and this would allow us to start to unlock again this kind of novel biology in these undruggable targets such as you know I on channels gpcrs and and and again going back it allows you to actually test hypotheses that you haven't been able to test before and and you know where the fields headed and and where I'm really excited is is ultimately being able to start to predict some of the biology so once you solve the design problem how do you figure out what targets we we should be going after that are you know ultimately going to give us the efficacious you know results that that that we want because we all know we can design an amazing molecule but if it's not to the right target it doesn't really matter and and so that's where we brought in you know expertise like on Andreas Bush amazing drug hunters that can help us take you know this you know technology we've developed and help us apply it to the to the right targets and the right biology and when you do that I think you really unlock I think really exciting therapies for patients Andreas you kind of jumped in later in the game then chanted at least on the upside side and in some would argue and say well okay sure AI is going to help here but wouldn't you rather be in an organization with all the vast resources that can kind of throw at this problem rather than go into a startup why why kind of you know skydive out of a giant mothership into something specifically focused on on kind of like data generation into AI and drug discovery rather than kind of just trying to make it happen in a large farmer company to be honest when when I was responsible for and yet a large farmer company AI was always a platform which I assume would take over number of aspects of discovery and development however there was at a time where a number of things were missing a the talent to join big farm I was missing the data in its quantity and quality were missing and also the complete understanding how do we connect all of those individual components in this complex aspect of drug discovery I mean charges mentioned that you know we are addressing right now the beast of basically generating the perfect antibody or biologic against the given target however it doesn't be clear it's absolutely critical that we have the right target and now if you think about the right target how much it involves to get the right target it again is a multi-dimensional aspect you know trying to translate data from a human being to an animal or back from a healthy to a to a sick human and and understand overall the the plethora of data you need to understand in order to go forward and then get to the point that a good target requires a good a good entity to to address it is clearly a complex aspect which is not really well covered by an organization by an or the organization of a big farmer where I believe you have from the one hand very good oversight in what is needed but on the other hand you don't have the deep expertise on the individual components and I think it's it's to me at least very clear that more and more big farmer companies understand that is deep expertise and how to generate an optimized antibody the deep expertise really going about signaling pathways understanding toxicology pathology and so forth you want to really dig into the experts which are very often biotechs which are very often outside of the own internal farmer commonly and and the farmer expertise that's interesting so so Andreas you actually just a paraphrase you actually think you can build because obviously farmer you know the industry looks soon thinks there's a lot of expertise there but is that what you're saying that sometimes it's I mean it makes sense on say deep AI expertise but even within biology you're saying you can actually find or build greater expertise in a certain application area in a biotech if I understood correct that's what I assume I mean I think what we have to appreciate right now not right now I think always is that the the game workplace is really at the end of talent game and we have to always understand you know what talent gets attracted by what type of background and organization and I would say that this very specific talent which you need to solve a very difficult question a very focused difficult question is more likely to be attracted by a highly innovative startup you know which comes with a proposition or hey you know if if if we are successful here we either gonna win the Nobel Prize over Generated Equity which is amazing to you both is attractive to two young scientists whereas if you look at who is attracted at the end of the day by big farmer it is people who want to bring drugs forward and translate somehow ideas but they also love the safe environment they always love you know the aspect of knowing you know if they build a house for their family you know they don't have to move the next month so I think you know it is there has to be in the appreciation of you know what drives certain industries and very often it is talent acquisition at least this is this is my big experience and if you look at the efforts the big companies have spent to build up internal expertise especially around AI how much they've invested compared to a number of biotechs it's amazing to see you know that still at the end of the day they rely very much on making partnerships for the right reasons even in the context of of absai there is a degree to which there is like so absai or my understanding of the approaches you can go for a a best-in-class treatment where the biology has largely been proven out but you can use some of the tools you have to deliver something that you know has longer dwell time in the body so you don't have to dose this refrequently or you know it's basically like for better binding affinity so it's like actually going to be more efficacious and then there's a almost separate kind of hypothesis that hey we can actually understand and and generate novel biological kind of like assets can one of those like BC's kind of succeed while the other does not or does the successive one indicate that the other is going to succeed and how do you think about kind of like strategically the approach and AI's call it you know potential to to lead really differentiated results in each of those areas. I can take this and then you know Andreas can comment but I think it comes down to a diversified portfolio and I think with every platform company whether it's you know AI driven or not you need to make sure that the first asset that goes in into the clinic one it's able to prove your your platform and your technology and and to the you know you ensure that the biology you know plays out and I think that we could have taken an approach where you know we take a bet on the the technology and we also take a bet on on the new biology but I think that that's too much risk and so what we decided to do on on the first asset to one a was you know we saw that you know the the IBD space it's a it's a very large market we started pursuing to one a before the Prometheus phase two read out and then shortly after Merck ended up acquiring Prometheus based on the the phase two proof of concept work and we really believed in in in that biology but the biology was de-rested was already in in the clinic but we saw a ways of improving the molecule you know both on potentially increasing the overall you know efficacy of of it but also increasing overall patient convenience and we we focused in on basically making a second generation TL1A asset that would have both of those you know attributes potentially and we believe that that's a great approach because again you de-rest on the biology side but you're able to to show a big win on the board from a technology standpoint and how your your technology was able to generate a differentiated asset now that's our first asset now if you kind of look at ABS 301 it's it's you know the complete opposite where this is a novel IO target it's novel biology we fundamentally believe in it we have some you know exciting results that are going to be coming out at our R&D day on December 12th on you know the MVVO efficacy but you know that's going to be obviously a lot higher risk you know moving forward then a best in class approach like on our TL1A asset and and this is where you you have to have domain expertise like Andreas who is an expert drug hunter I mean over 10 drugs approved to really look at the the pipeline and the targets we're going after and being like are these good targets to go after both from a best in class perspective as well as you know first in class and and so getting the targets right whether you know whatever strategy you have is is super critical and I think it can make or break a biotech company and Andreas I mean please weigh in I mean you know better than anybody you know how important you know getting that the target selection is as well as the right strategy and the right blend of first in class best in class for sure I think you took away my thunder already so well let me let me maybe add that I think this is the most decisive aspect of most farmer companies fighting exactly you know the right balance how we structure portfolio and and in general you can distinguish between those two aspects the best in class versus the first in class and they all come with different risk if you look for the best in class it's pretty obvious that the one number one technical risk which is that at the end of the day starting from a target to approval you lose about 99% of your approaches or if you now go to a layla phase in the R&D discovery value chain proof of concept you know if you generate a proof of concept you really avoid about 80% potential so you really have to think of best in class of I take away a lot of the scientific risk if I go really with a target which has shown a proof of concept in patients just think about it it means oh it got through toxicology got through CMC issues it showed and that's the most important aspect and the appropriate profile in patients which makes me believe it's gonna be a drug for the first time for the first time not before so you really cut a risk dramatically well having said that why why wouldn't everybody do that well obviously you add commercial risk you add the risk you know that your number 75 so you have to exactly understand okay where is the sweet spot where is it really where best in class means a very relevant difference to the patients which is appreciated by healthcare providers which is appreciated by a very embarrassment so you have to find exactly that sweet spot where you are among I would say depending on the market size the top three four companies where you really know that you can differentiate from the first in class in very meaningful parameters at the same time like Sean said you know if we have the chance while building up our portfolio you know with the first asset and the second asset as a best in class approach if you have the chance of getting first in class with a somewhat duress approach of course you will end up with a highly innovative but more risky target and this is what we're doing with our reverse immunology platform where we are in a situation of yes we come up with targets where nobody else but as far as we know it would work on right now which however are already somewhat validated because we start off with a human antibody identified in a tertiary lymphoid structure and therefore you know we are much further away from the typical oh I find this type of overexpression here and you know hypertrophic mouse displays this here so you start off at least at the beginning with a somewhat higher validation then you usually would does it mean it be risks the project well to zero extent but it still is much much higher risk much much higher risk than the first approach on the technical side not on the promotion side of the growth yeah can you talk about the specifically to let's start with best in class like that approach how does AI actually help you there or how does what you're doing actually get you to the ability like I said it's like oh that sounds like a great idea I'll throw some money at getting there well I can't I get there but you can like what are you doing that's different from how it is traditionally done that the yields kind of like that better binding efficiency and 12 times in terms of you know what we're able to you know deliver on I think it's you know three threefold I think first it's differentiation so being able to you know let's look at you know two one eight for for example with with that there were some known liabilities with the known competitor molecules that that were out there you saw you know with the the relevant molecule you saw increased ADA response when compared to the Prometheus Merck molecule and and we saw that that was likely B cell mediated so essentially where the antibody was binding to the target was leading to immunogenicity and so we took that into account when designing this molecule and we we essentially were able to design a antibody that bound to an adjacent epitope to the Merck which we believe was a conserved epitope that would decrease overall immunogenicity when compared to the Roy van molecule we then engineered the molecule to to have high potency so binding with with with high affinity and we're able to show that we could have differentiation there and then one area that we see as additional differentiation could lead to potential increased efficacy is being able to bind to both the monomeric version of Tijuana as well as the trimeric version and what you you see with the competitor molecules is that you either get you know binding to the monomer and no binding to the trimer or very you know weak binding to the to the trimer and we believe by by using the the AI model to to bind to both of those versions we could potentially increase overall efficacy and so that's how we use the and I guess the last piece I forgot to mention was increasing overall half-life so being able to engineer the molecule to stay in the body longer so you have less dosing frequency so going towards kind of patient convenience so instead of dosing once monthly you dose once quarterly so that's how we're we're using AI to create a differentiated asset now the other you know area that we're using you know AI to you know to to accelerate this and and we really believe to kind of break the paradigm of how the economics about tech traditionally work is we were able to use these AI models to rapidly get to to a drug candidate so we're able to get to a drug candidate in 14 months you know normally that could take two to three years to get to to a drug candidate we're going to be in the clinic with this asset within 24 months so early next year we'll we'll be in the clinic normally that takes five and a half years to get into the clinic so we're cutting the time in and half and then additionally we we invested roughly 13 to 15 million dollars to get this asset into the clinic traditionally call it 50 to 100 million dollars it would take to get a normal asset into the clinic and so we're able to completely change the paradigm and and economics of how traditional biotech has worked and and so you're able to get more such on goal with the same amount of investment so instead of investing you know 100 million dollars into you know one asset you could invest that 100 million dollars into five to ten assets and that doesn't even take into the count those assets you're developing are more differentiated which means that you're likely going to have higher probability of of success in the clinic and so that's how from an economics perspective we believe AI is going to completely disrupt how R&D is is done these feel like a very important you know massive accelerator and catalyst of overall to the industry right and so for companies like upside in particular that we study and think about a lot because if you look at you know if you if you take the view that we we didn't really have what a medicine say 100 years ago you know by some measures we've made great progress but but also things you know move way slower than we like and particularly things are much more expensive than we like and if you you know if you took out the institutionally university funding the grand bodies and all of the bus five attacks along the way that didn't get acquired you know I don't know if anyone's ever run the numbers it's pretty tricky one to make but arguably drug R&D historically has barely wiped its own face overall as as an ROI you know compared to other other sectors and I think you know some of that's filtering through into what we're seeing in the public markets right now with this historic number of negative EV biotechs on the public markets and I think you don't have to increase efficiency and speed or you know decrease cost that much to really tip the scales and favor of this is a good thing to do and I think with the kind of speed up you're seeing you know that becomes very exciting is it possible to double click a little bit on how the AI actually does drive progress at different parts of the pipeline because I think one of the things if we look at you know the GPT revolution one of the things these models are very good at doing it's I think one of the reasons we're very excited about absolute technology is modeling sequences is a very good use of these things modeling complex non-linear interactions in a very high-dimensional space it's difficult for a human to model and we see that in natural language but you know you seem to be at the forefront of doing that within biology you know if you're going to understand you know double click a bit on that and talk about the challenges and the aim able is for that to actually to deliver molecules ultimately yeah so essentially like what we're doing with these you know models is we're first off you know training on sequence-based data so data that's coming from the ASA that we had developed we're able to multiplex you know antibodies you know binding to targets of interest and looking at what epitope is a binding to what affinity and then additionally what we're able to do is also leverage you know publicly available data on the structure of protein protein in interaction so you have a structural-based component as well and essentially these go into various different models like you know diffusion based models graph neural nets and you know we've gone to the point now where we can take a target of interest we can specify the epitope or the location that you want the antibody to bind to and so the again the input is the three-dimensional structure so you have to have the structure of the target for the model to work and and that can be derived experimentally through like cryoEM or extra crystallography or you can use something like alpha-fold as the input or the structure from alpha-fold as the input and from there again you specify where you want the antibody to bind and then the model is unable to design the CDR so think of the CDRs as almost like the fingers that grab onto the target of interest you have six fingers that are important for binding to to a target and the model is able to design those CDRs that combine to that particular epitope of interest now this may sound really simple but if you think about how things are currently done you have no ability to design an antibody to hit an epitope of interest this is the first time ever scientists have the ability to design an antibody to an epitope of interest that has the affinity that they want because if you look at current approaches one of the the most you know kind of recognized ways of you know creating antibodies and it goes actually back to you know one of the you know great biotechs that we have today re-generon re-generon you know created the the humanized mouse to to generate antibodies so essentially what they do is they take a let's say cancer canter antigen of interest they inject it into the mouse and the mouse uses its immune system to generate an antibody that binds to it to that target and then they take those those antibodies and they screen and figure out which of these gives me the the ultimate biology of interest but the issue that you have with that is you have no ability to control you know how the mouse immune system generates the antibody so you can't tell a mouse to generate an antibody to a particular area of the target the mouse gives you what the mouse gives you and so now we're able to use AI to design the the antibodies to have all the attributes that that that you want you know whether that's to hit this particular you know epitope to to you know create an an agonism you know antibody or antagonist or have a particular you know affinity that that you want to test again these these hypotheses that you you haven't been able to to test before and I I think a great example of of this is actually our partnership with with AstraZeneca so we you ended up closing a partnership with AstraZeneca late last year started on the program earlier this year and and this is a program that is is focused on a oncology target an oncology target that has you know been a struggle for for them to produce antibodies towards and we we identified six epitopes that we wanted the the antibodies to to bind to and again there are no known binders to to this target and we were able to successfully design six antibodies or we were able to design antibodies to be six epitopes and this was a huge accomplishment because again this was a target that you know folks that struggled to to drug and and you now have an AI model that can generate a drug candidate to this really exciting you know target that's a huge that's a huge breakthrough now whether or not the biology actually pans out it you know is is is a different point but the point is that this has the ability to completely change the the the game on on how we think about the design how can how can we you know more rapidly test you know these drug candidates and so I think it's a I think it's a huge breakthrough on being able to have a model that is you know epitope specific again testing hypotheses that you wouldn't be able to test before and really solving the design problem of of antibodies is that same model what I'm going to get to you in a second Andreas is that same model does it is it usable in the kind of like antibody creation optimization space and in the kind of hey we're going to go after novel biology space or or do you have different approaches to those so we have two two different approaches and and Andreas can can touch on this so the the this is more hitting on the design using the this AI model to design the antibodies for let's say a best in class strategy but then we have our reverse immunology platform which Andreas was touching on which allows us to discover these new novel targets to go after so they're two separate platforms that that we have here one is you know design of antibodies and then the other is is target ID very little to add maybe maybe I want to add a couple of aspects which Sean has not mentioned on top of the once he has mentioned which is doing it that way you know generating the novel binder we also end up with an incredible diversity of the CDR region which gives you a complete new playground for generating IP you know because it's not just the one sequence of maybe plus one minus the minus it for which you can generate IP but you can get you know opening basically the entire CDR you can get protected because you we saw you know if we compare to to Tristusema at our her two binders which we generated we saw binders you know with a distance to the original Tristusema you know with the distance of 9 to 10 to minus it so we could exchange 9 to 10 to minus it to to still get a very potent binder so so I think that is one of those aspects you know which is in the context of value generation so very important you know that once you have generated this antibody the chance is very very high that nobody can touch this IP or do do a little change that's that's one of the things which are I want to add then yes we are working very hard to fuse on the one hand but the no more model to generate the antibodies with our model which at this point is also separate to optimize the model tunes it still goes pretty fast you know within a few weeks we can you know apply the model to optimize the the antibodies from the first binders we get and then again is very very important to keep in mind how is it done traditionally versus what we can move we can address aspects like affinity, offline, ADCC we can address pH dependent binding and so for all at once doing a multi-dimensional optimization and we know exactly what we want to do that is very different to the mouse I always tell a show yes you can tell the mouse what you want unfortunately the mouse doesn't listen to you and the mouse you know spits out what it spits out so we can do that and and then the final aspect which wasn't covered yet it's nice to be fast it's nice to be fast but why is it so nice to be fast it's not just because of the costs you know I mean working on the project for a year really costs you less than working on it for three four years but you have to again keep in mind the competitive event to gain out of that you know a year in the present R and E value chain is gigantic if you keep in mind that everybody with the present methods is working on the same targets everybody you know is so when I say everybody I would say there's always three four competitors you know working on the same target one year is not a little bit more or a little bit less of the value it may be digital it may be all of the value of zero of the value so I think we we have to keep in mind what time really means and what a year means it doesn't just mean if you have a blockbuster that you earn one more billion at the end of the life cycle it may mean that you get it all or you get nothing can I unpack one thing you said so you were describing essentially if you imagine you have like the spot you're supposed to plug in on the protein here and using your methods you discover a spot that's maybe over here and then you can IP protect like essentially everything around the derivations of what you're doing such that it would be hard for somebody to be like oh yeah that looks like a good spot to operate in and be like okay we're just going to change one amino acid and actually land there because you've like actually IP protected the position much more completely is that the right way to think about it yes that's usually where Sean Johnson but yes that's more than what I want to try to say if you look at for example the Trastuseum up you know CD up 3 which we have the noble design you know it's 30 in a minor assets and we come up with binders where we exchanged 11 of them and they still you know produce a good binder so you know there there's no room left we need to search for any changes you can make yeah to put it in perspective I mean that there's more sequence variance in an antibody than there are atoms in the universe and so it doesn't matter you know how robust your your screening platform is you're just never going to be able to to you know screen that many antibodies but what you know these these AI models are allowing you to do is actually start to explore that much larger search space and you know what's you know we find actually quite quite interesting now especially after the the amgen sonofi case which you know really showed that you need to show enablement in order to get broad claims you know I think you know the amgen was trying to get a whole target claimed based on you know 14 to 20 antibodies I forget but it was a is a very small amount of antibodies about to this target they wanted to get you know broad broad claims and the us you know PTO came came back and said that they didn't have enough examples to issue a broad claim towards towards that target but now if you can actually use the AI to you know search that whole search space and come up with millions of designs towards that given target and then not only that you have the wet lab to go then screen those those million antibodies you now have enablement to get much broader IP claims then you could have ever gotten before so you actually need both and I think that's one of the things that we didn't we haven't touched on yet and I think that that's one of the big reasons why we've had the success that we've had is and and you know Charlie and I've talked about this a ton is is this this active learning loop of being able to generate data in the wet lab a physical wet lab and you know feed that data into your model train your model and then be able to go validate that the model see how accurate that that model is and and we're able to go through that active learning loop right now in in in six weeks and so we're able to generate billions of protein protein interactions to train our models and then we're able to you know look at three million unique AI generated antibody designs and validate them to see how accurate that the models are I mean that that throughput as well as that that cycle time of of six weeks is is really the best in in the industry or the best of what we've seen and and we're constantly looking to drive that cycle time down because the faster we can drive that cycle time down that six week time period right now to you know not not just weeks but to days that's going to allow us to dramatically increase the the overall accuracy and predictability of of of these models and and Andreas talked about kind of two separate models we have the de novo model that's like your global search so you know basically you know you you want to hit the country you want and then you have the local you know optimization you know you hit the city that that that that you want but ultimately as you get more and more data these models actually start to to to you know converge and you can hit the city and and the right country at the same time and so you can get both like let's say the efficacy of the model or the the of the molecule plus the developability and manufacturability all in all in one go but in order to get there you got to get the right amount of data and you've got to have that active learning loop and and and that's what's really led to the success of that that we've had and and we wouldn't we wouldn't be more right if we didn't have that that that active learning loop at that six week time period definitely seeing a rise of people talking about you know good biotech operating at the beach right where wet meets dry um and then you know we think the next level is the kind of active learning that you're doing and you're handing over the cars to the sort of robot scientist and maybe eventually the founder can go and sit on the beach but but those days are probably still a way a little ways out I don't know exactly what I would do on on on a beach though at at that point in time dream up the next biotech maybe I mean that this sort of leads into a couple of other questions I had about first of all you know thanks for highlighting some of the combinatorial complexity you deal with because it's just fascinating in biology how quickly it blows up someone told me a great start the other day that I did it I did objectively is roughly accurate if you spent your entire life doing nothing but reading your own three billion base per genome at the rate of one base a second it would take you your entire life like a century almost exactly and and that's just that one genome with with no variance um so fascinating and of course it blows up very quickly and I think it's one thing just for our listeners as long that you've you know identified but it's not necessarily super obvious to people from outside of bio is that as you've said function essentially is modulated by structure in biology whether that's you know antibody structure drug structure I think and that's in the case of proteins and antibodies modulated by sequence so you're operating up at the sequence level where AI is really good and training on the outcome which is the functional output and I think that drawing the line between those things is super important and I guess my question was how much biology is there like as you sort of in the limit as you as you as you as you sort of look out five years ten years 20 years how much of the current biology is understood that will form those drugs of the future like where where are we on that kind of exploitation exploration i.e. first in class versus first in class and in terms of the future pipeline the humanity has to play with what do you think we are on that journey yeah absolutely i mean you know where where we're at right now is like we we've built scalable biology for you know predicting protein protein you know interactions and you know now we're we're looking to scale data is is on the the in vitro side and and the in vivo side and if we can effectively figure out how can we scale the cell-based assays and and how can we you know scale you know you know animal studies it will at least have a a a model that can mimic animal studies you know very very effectively that's when you can start to you know model the the biology and and start predicting you know what what targets to go after not only like what targets but what epitope on that target should you be hitting to give you the overall efficacy that you want to how do you ensure that you you get you know antagonist antibody versus a agonist uh you know antibody and i think like all of these uh pieces i i do believe that we're going to solve but we do need the the the data in in order to to inform these models for for these models to be accurate enough to to really start to be predicting the biology and so that's kind of a big focus for for us but i do believe in the next you know five plus years i think we're going to be at the point where you're able to have these these AI models you know actually start to recommend you know epitopes you should be looking at recommending what targets to to be looking at in order to achieve the biology that that you want for for a given you know indication of therapeutic area so i think that that's where where everything is is headed i don't believe that that we are we are there yet on on actually using AI to predict the exact you know target and getting the right mechanism in in place and you know having great success of of having that pan out in in the clinic i think we are very much getting you know towards that that direction um but i i think there are you know some some limitations there but i do believe that that's where we'll be in the next like five five plus years oh no it it just it's incredible just how fast things are progressing i mean i i i i i i i i i it could happen in in in you know in two to three years versus five under i said yeah under that circumstances is the portfolio mix just hugely biased towards first in class as opposed to the best in class or do you think you're always going to have kind of like a mix is it's going after best in class just a way to de-risk the platform and ultimately you think that abseye will be targeting hey we're we're discovering new biology and we're targeting it or do you think it'll kind of blend right now our front runners are best in class um and and certainly you know with the next generation of assets we bring it to our portfolio we try to have a somewhat de-risk first in class um and i guess what much i'll try to indicate is that at the end of the day uh of getting early targets before you have any that idea of you know how an entity against it really works uh de-risk um you need tons of data and this is what is happening right now i mean if you look at the data generation as we see it right now both on the human side so we get i mean look at just all the keyword studies we we get which we can analyze with potentially very soon you know artificial intentions uh about all other genetic data we get about all the all sorts of disease data we get at the at the really high scale of uh uh of quantity and quality that all will accelerate um but again we're not there yet but that all will accelerate understanding better human pathophysiology and the value of an individual target yes we're not there yet very very clearly as by the way whatever we talked about today we want to further improve whatever we talked about today we're not there yet but we really feel we are on a good way to get ultimately where we want to get which is you know we have an idea about the target it gets validated very quickly and then we really can generate one hit of a button uh and optimize antibody the novel and can put it into the right models to test it biologically this is the idea and the vision we're living for and working our butts off right now um and and you know it's difficult to exactly say when for a hundred percent of all targets we can deliver that always keep that in mind today we know there are targets around where this works already but what our ambition of course is is we want all targets to work we want especially the targets to work which are you know historically undruckable you know all those things we want to address with what we're doing right now and i think we're you know we have a good start and we are on our way but what we can say we're not there yet but we is soon we're going to be faster there than we actually think because if you look at what the contribution of AI was and how things developed over the past five years if you look at the interview of David Baker it's very clear that most of the advanced events events has really happened over the last five years because you got this combination of great scientists getting together with the right models technologies uh building on each other but the huge 12 of data uh being generated on the other side and and i think this is why why it's exactly it's not so easy to predict you know how we're going to be there by the end of the year or our next year but i think uh overall we're optimistic that we're there much faster than probably uh probably like well we're definitely as investors of course uh but fellow entrepreneurs and and certainly just for society and there's people with families and what have you were cheering you on you know super exciting and important to drive better outcomes from the immune system modulation especially because you know there's there's great data actually the huge amount as you you all know well a huge amount of aging is actually immune degradation finite evolution leading to everything from obviously cancer but also infectious diseases and various other things and you know we're seeing increasing evidence of contribution to even neurodegeneration and a ton of other diseases that neither the lack of understanding and the lack of our ability to modulate the immune system is led to one that is just made you know so much more is known about aging and i tell you nothing nothing is closer to my heart than thinking about mechanism to act on which can clearly in help enhancing a healthy and extending a healthy longer life uh but now we come to the next problem which is translating a wonderful mechanism with a wonderful compound into an appropriate clinical study so charzia could help me there how i would study extension of life in a way which would be accepted by regulatory agencies anyway sorry sorry i just i feel uh put podcast 2.0 the the sequel is a little cliffhanger i mean look i think there's some interesting work that is going on and could be done around a panoply of market as sort of aging markers but including things like the whorevast clock right the methylation clocks which are looking like reasonable proxies though with their problems right not least of which that they they shell tight is a bigger driver of those but i think like those now those problems are being understood maybe that's where i used to understood them and even even a different age range of things like grip strength and ability to walk far i think there's a bunch of stuff right that ability stand on one leg there's a kind of a comedy biomarker but it's quite a good one and there's a bunch of things that you know i'd love to debate on another podcast because it's definitely an area of close to heart 2 just to close out though maybe you know join i think one of the things that really impressed me about you as a entrepreneur and outsider company was effect i mean we'd met each other before but effectively bumping into you in in the maelstrom of nurex the machine learning conference where i think you probably were uniquely the only CEO of like a half billion odd market cap company standing there defending and presenting your company's technology and you're thinking to essentially a crew of phd students at your own poster which i thought that's just so cool and so rare to see both the level of knowledge to be able to do that the willingness to do that and just the excitement for the technology could you talk quickly about your own journey as an entrepreneur and maybe i mean you're in your mid 30s you've already achieved an amazing success with what you're building and maybe advice for your even younger self of maybe a decade ago what have you it's really about being a lifelong learner and and it's checking the the ego at the door and and i think it's it's recognizing like what you you don't know and and and what you do know and and really you know surrounding yourself with people smarter than yourself i might my dad taught me that that lesson that at a young age is like you know higher people much smarter than than yourself and and to do that it means checking checking that ego at the door and and recognizing like you're not going to be the domain you know expert on on everything and and that's okay and you know hiring people you know like like on Andreas said that have come on you know you know folks like you know Zach Jonas and who's our you know CFO and CBO he was like the you know the venture partner the first venture partner that put you know venture money and into you know abseye you know folks you know even like Shelby Walker we just you know you know brought on as our as our head of legal i mean i have learned so much from each of these you know individuals and and it's i think allowed me to become i think a very well-rounded CEO and and you know it's allowed me to continue to stay curious i think curiosity is as what has has driven me and and you know being surrounded by these these really smart people it kind of helps me you know look at things from a different perspective and and at the end of the day and i think it helps you know make sure that you know the the vision of abseye is crafted in the right way we're going and we're rowing in the right direction and and it's kind of being able to kind of synthesize all the different you know inputs that are coming in and being able to say okay hey this is direction we need to be going and and here's where we need to be like doubling and tripling down but if you don't have that you know you know information you're not curious you're not surrounding yourself with with people you're you're not gonna make the right decisions at at the end of the day because like you just don't have that that information and and so i think uh to me it's it's it's stay curious uh higher people smarter than than than yourself and just be a lifetime learner well on that note uh i hope that the listeners of this podcast learned something from this conversation i certainly did uh and i really appreciate the two of you coming on and we uh hope for the best for abseye and think that you know generally i think it's amazing how quickly the AI space is moving and there's probably no higher leverage space than biology to deploy it against because it's so data rich and we really have like the only the thinnest layer of understanding on top of that data is to what's going on and so this seems like the way to find out and then deploy that data against you know real things that'll make a difference in people's lives so thank you yeah thank you for having us this has been a ton of fun thank you for listening to this episode of fyi the four your innovation podcast if you enjoyed this episode please feel free to leave a rating on whatever platform you're listening to and if you haven't already please make sure to subscribe to the show and follow us at arc invest on your favorite social platforms so you stay up to date whenever we drop new episode arc believes that the information presented is accurate and was obtained from sources that arc believes to be reliable however arc does not guarantee the accuracy or completeness of any information and such information may be subject to change without notice from arc historical results are not indications of future results certain of the statements contained in this podcast may be statements of future expectations and other forward-looking statements that are based on arcs current views and assumptions and involve unknown unknown risks and uncertainties that could cause actual results performance or events that differ materially from those expressed or implied in such statements