Digitalizing The Biopharma Ecosystem With Kat Kozyrytska

In this episode of “Better Biopharma,” host Tyler Menichiello is joined by Kat Kozyrytska, a technology advisor and portfolio strategy leader. Kozyrytska shares her thoughts on how technology companies fit into the broader biopharma ecosystem. They discuss how AI is blurring the traditional boundaries between tech providers and therapeutics developers, as well as the increasing shift of tacit knowledge from human minds to the cloud. Kozyrytska also weighs in on how biotech companies should think about building, buying, or collaborating on AI platforms. 

Keep up with everything biomanufacturing at Bioprocess Online: https://www.bioprocessonline.com/

Follow Tyler Menichiello on LinkedIn: https://www.linkedin.com/in/tmenichiello/

Subscribe to our newsletters: https://www.bioprocessonline.com/user/edit/subscriptions

Tell us how we can improve. Take our 60-second survey: https://docs.google.com/forms/d/e/1FAIpQLSfZGJluCmEhm1MBQvcm7_gtBEXuTx5v42LJ1mx9bCUKo8fwfg/viewform?usp=header

Transcript

SPEAKER_00 0:17

Hello and welcome back to Better Biopharma, the official podcast of Bioprocess Online. I'm your host, Tyler Manichello, and on this episode, I'm joined by technology advisor and portfolio strategist Kat Kozaritka, a longtime friend of the show who I've worked with on past projects around digitalization and technology adoption in biopharma. Today we're going to talk about the critical and undervalued role of technology companies in biopharma and the necessary cultural and structural changes needed to enable the industry's digitalization efforts. Also, we're going to talk about AI and what therapy developers should own, outsource, or solve when it comes to enabling their development and manufacturing needs. So, Kat, thank you so much for being here. And for those in the audience who might have heard me say undervalued role of technology companies, they might be thinking, what are you talking about? I think they're pretty valued in our space. So, Kat, if you wouldn't mind, please give our audience a bit of a background about yourself and explain what we mean by undervalued role of technology companies and how do you see their current role in today's ecosystem compared to the past decade? How have they evolved?

SPEAKER_01 1:19

Thank you, Tyler. It's great to be here. I really appreciate this. Um, so just a touch about me. Um, I uh spent my career on the technology side working on portfolio strategy and commercialization, and especially at times like this, um, that we've encountered in the past, translating market technology um regulatory shifts uh into business action. Um so um uh in terms of the question that you asked, uh uh what how how do we perceive technology side today? Um, I think uh in our sector, which is structurally quite different from uh other uh sectors, uh, is that the situation is that we really therapy developer-centric, um, which is understandable. Therapy developers make life-saving therapies uh for patients. Uh, but we often forget that in fact the sector is an ecosystem of companies. Uh and that's the part that's really changing now, uh, in part uh because of the emergence uh and just such centricity of AI that we're living through. Uh and so um I think historically what we've seen is that um you had, let's say, capital equipment, right? We have a box that does something, that puts some measurement maybe, um, and there's a company behind it that makes uh that box. But and I'm simplifying, of course, right, the different kinds of technologies. Um, but where we're moving to today uh in this um space of emerging AI technologies, but that's really impacting the entire ecosystem, um, is that there's a lot of knowledge and know-how that comes very close to therapy development already on that, what we would conventionally think of as tech side. So almost the border between tech and therapy development is becoming more blurry. Uh, and we've seen just a little bit of this in the past. Uh, we've had therapy developers acquiring technology companies. Um, and so that is an interesting few instances we have historically, but now um that's happening just a completely different scale.

SPEAKER_00 3:38

Great cat. Yeah, thank you. I I agree with what you're saying about looking at the industry as an ecosystem and understanding the need for technology companies' roles to evolve. And I'm curious how you think not only technology companies, but also therapy, therapeutic developers and manufacturers, how do all these different players and their relationships with each other need to evolve in the coming years to really enable the next generation of biopharma in a way that can compete with other industries who can move faster because they're less regulated?

SPEAKER_01 4:11

I think essentially AI is serving as a kind of bridge or you know, forcing that binding between the different parts of the ecosystem, intertwining in a way that is much deeper, uh, much more profound, um, and probably would benefit from some overarching strategy on the part of every one of the players. Um, so on the part of the therapy developers, thinking about what should they own uh and what should they where would they benefit from somebody else developing the technology? Uh and then you know, on the C DMO side, right, as they're as they're looking across, um again, the question becomes who owns what um within that ecosystem when it comes to AI models and broader AI learning. Um, and I think the way that that's a shift for us is because we've been much more, let's say, independent. Uh you, of course, there's voice of customer, we've tried to stay connected, we go to conferences, which have been great in terms of the information exchange, but there has not been a pressure to be so interdependent. Uh and uh with AI, because AI is getting trained on data and that data sits and is owned uh by so many different players. Um, if you want to have a model that actually learns from some end-to-end, then that there needs to be a plan uh for how the different pieces of data will interact, how the different players will interact, and uh who will own what in the end. And there is already some um discussion and momentum, of course, to really figure out the ownership of IP uh between the different players within the ecosystem. But what we're seeing is the divergence. So in this case, the technology company will own maybe this bigger piece, and the therapy developer will own a smaller piece, but then in this very other, in another case that's very similar to that first one, the roles will flip. So essentially, there's no top-down plan for how to split up that ownership, which makes sense. It's a market, there's no governing authority that would dictate that uh you know that we ought to go uh a certain way. Uh, but in the end, uh we want to arrive at an efficient and optimal market level solution. So um that will likely require some strategic thinking on the part of every player that they pick a lame for their development and in the end, ownership where they're best positioned to uh both develop and own in terms of the resources, the data, all the other pieces that they have.

SPEAKER_00 7:09

Yeah. Thank you, Kat. That's great. And you touched on something that I wanted to ask you more about, which is that you describe this kind of transition of knowledge and data from which we've kind of witnessed over the past couple decades, I'm sure, but especially in today's day and age, the shift away from knowledge being stored in human minds to a unified digital ecosystem that could be sold, scaled, and reused in tools like AI. And I'm curious if you could talk a little bit more about that transition and how it's kind of affecting the way, as you're describing, that companies need to think about their data, manage their data at an infrastructure level. And in today's environment, what do you think gives biotech companies a competitive edge in terms of their AI integration feeding into that data?

SPEAKER_01 7:55

There's a lot in your question, so I'll unpack bit by bit, and then you tell me if I don't get to a particular part of your question. Um, but in terms of that information storage, um in large part because we're such a hands-on process, at least today, right? There are a lot of humans involved in both the research side, beltman side, and manufacturing side. So a lot of the information and the know-how is stored in the heads of the employees. Um and um, of course, with uh automation and digitization, um, the transition is such that we're trying to transfer the knowledge of the humans into the robots, the machines that are uh trying to execute on these processes. Um the kind of intermediate state that we're in now is that the you know, we have we're trying to develop supportive digital technologies. Let's say before we even get to the robots doing the process, can we just summarize the knowledge into some kind of a, let's say, again, simplifying, but model? Uh and of course, the challenge with that is partly philosophical. If you have a human who's working on the process, um, do they own that knowledge? Does the company own that knowledge? If they've been there 20 years, maybe they came in, they learned something in the beginning, and maybe that's owned by the company, then they worked on something, they developed it. So there's there's a lot of questions that will come up. Uh, and this is not just in our industry, but really uh across industries to figure out how to attribute and distribute the value uh of the knowledge. But the other really interesting part for us um in science generally is tacit knowledge. Uh, and so uh there are many books written about this. Um, I think my favorite one um gives an example of how this is for it's a massive uh physics experiment. Um, so they thought they understood how it works, and then a guy left and it didn't work anymore. So they went really digging into why, why it stopped working. And so it turned out that at some point in the process, he would put a thread behind his ear, get the wax onto the thread. And so that changed the conductivity or whatever, right? I mean, it's some very advanced things, and that changed the way that the experiment ran. Uh, he did not register that he was doing that, he did not write that down somewhere. It's it's just sort of gass it. Uh, and for us in uh in life sciences, of course, there's so much of that. Um, you know, for example, I don't really need a machine to look at or to measure the protein solution to see how much or it's to see whether it's aggregating or not. I can already tell and I know. Uh because, you know, I'm at eight years that I've spent in a lab, I've walked at a lot of tubes. I know what they look like when the protein is starting to crash out of a solution. Uh, and so that's just after eight years. I mean, imagine you do this for 30, 40 years, right? I mean, you get a really good understanding right away without the kind of profound need for 15 different machines to measure things, you already know. Um, so I think registering that kind of knowledge is going to be quite difficult. And transferring that kind of knowledge into machines also quite difficult. Of course, we've been on a path to that. Um, but you know, to give another example, somebody again, just eight years of experience, but you walk into the lab, you don't need to test for contamination, just by the smell, you can already know. Uh, and so again, it's it's very difficult to record that kind of uh expertise into a robot. Now, are we perfect in that sense? No, we are very limited. Uh, dogs smell better than we do, cats hear better than we do, shrimp sees broader um spectrum of light than we do. I mean, just all kinds of things where we're imperfect. Uh, but we're kind of an optimized uh mechanism for multi-sensory detection. So, you know, as we try to automate the processes, um, can we build a robot that's as good as or better than we are? Um, and then I think a second very important question that often gets missed is on the company side, let's say a therapy developer, right? Does it make sense to have that robot? Is the energy consumption, the tokens, the water, et cetera, et cetera, is that actually cheaper or in some way more reliable better than getting the human to do it? So I think the very first question is um, are humans better at some things and should we then do that? Because it's it's a cheap and fast way of doing it. Uh probably not so consistent with that vision of uh, you know, 4.0, 15.0, uh, where we have all automated robots, but the reality that we live is that humans are actually quite optimized. So replacing them in some situations is just going to be tricky. Uh but go getting back to the question of uh transitioning from human-based knowledge to um AI or machine-based knowledge, with humans, yes, you can hire a human from your competitor, and you know, the industry has done that. And so then you get some diffusion of knowledge that way. And arguably, this is actually really good for the industry because you get to bring up everybody's best practices. Um, maybe even when they're not spelled out in industry, um, let's say uh industry bodies, right? When they're not spelled out in a manual, you still get some exchange through that flow of humans. But if you're now recording that type of knowledge into an algorithm, there's an opportunity to take that as a very practical, implementable industry best standard and take it across all organizations so that everybody lifts. Everybody has that best uh practice that they start with. They don't have to arrive at it by themselves. Um, so again, big picture could be really good thing, but that brings up the question of who owns that best model? Is that a shared um industry knowledge? Is that owned by a technology company? Is that something that you know therapy developer creates within themselves, but then holds really tight? So I I think there are just going to be really interesting shifts um in in terms of um that knowledge sharing. But the core part is that if you can digitize it, if you can make it into an algorithm, then you can crystallize it outside of the humans and outside of perhaps even the therapy developer who produced the data upon which it was trained and so on, and then you can sell it to someone. So, because of that commercial aspect, I think the transition will be uh a very interesting one.

SPEAKER_00 15:23

Yeah. Thank you, Kay. Yeah, I did cram a lot in that question, and you certainly checked all the boxes in your answer. I appreciate that well thought out and um articulated answer. And you raise a good point about ownership uh of the data, and I mean that can kind of lead into our build by or collaborate part of the discussion a little bit. But taking a step back from that, I'm curious to hear. Um, you know, this show seeks to serve therapeutic developers ultimately, the people trying to make these medicines, develop them, manufacture them, bring them to market, um, and specifically on the development manufacturing side. I'm curious as it relates to what we're talking about with um structuring data and collaborating with technology companies and and just developing these medicines. I'm curious what would your advice be to smaller biotech companies trying to stay ahead of this AI curve and and use it to their advantage? You know, what's what do you think is the single most important thing they could be doing to prepare for the direction that the industry is going, which is hyper-digitalized, perhaps shared data, perhaps just I I think ultimately the ethos of where we're going is reducing silos, right? That's always that's been a conversation for before I even broke into the into the space. So I'm just curious to hear your advice. Uh what what's the most important thing that biotech companies can be doing right now in this ecosystem to get ahead?

SPEAKER_01 16:53

Uh lot of pressure with the one most important thing. Um so I'll give me three talk around that.

SPEAKER_00 17:00

I'll take I'll take three if that makes it easier to answer.

SPEAKER_01 17:04

So uh I think maybe I'll anchor to your to the part of your question that is about silos. Um and I agree that we have been very determined to break down silos, but the question is which silos? And I think that varies depending on which part of the ecosystem you look at. So um, you know, if you're a large um manufacturer, then you probably have multiple sites or multiple teams that are running uh within each of those sites. So you want to break down your internal silos. Uh if we look a bit broader within the ecosystem, um imaginably that there are parts of the ecosystem, let's say, you know, maybe we can zoom in on CDMOs, uh, there they have visibility into part of the data set, but almost never do they see clinical outcomes. So for them, probably breaking down that silo would be tremendously helpful in terms of developing better processes. Um that again, we have to we have to ask who will own uh that knowledge. But but setting the ownership aside, I think we can all agree that better processes across the industry would be better for everyone involved, including the patients. Uh and uh then, you know, of course, on the let's say, instrument, uh, media, reagents, um, and software services, uh, then similarly want to see more data. So maybe breakdown fragmentation between companies uh as well as within that uh workflow. So um, yes, uh the breakdown of silos is key, but uh we we have to understand which silos. And then back to the question about what's important for uh small companies, small and medium size. Uh there was a really interesting analysis um by someone whose name, unfortunately, I'm not gonna recall at the moment, uh, but I think uh they were interviewed by uh Endpoints News about um the XPI index. And so the really interesting part of that analysis to me was that the size of the index doesn't seem to change very much over time because as small companies develop their uh pipelines, they get bought by large companies. And in part, at least, and this is my opinion, this was not part of the analysis, but um, it's that small and medium-sized companies are less well set up to manufacture within our sector than let's say large players. So uh maybe have a great uh discovery concept, but then when it comes to manufacturing, again, because we're uh limited in the extent of industry-wide knowledge, we can imagine maybe you hire somebody who has deep knowledge expertise. But now we imagine a different future where that uh development and manufacturing knowledge, there's some baseline of industry best practices that you can import. And so that sets you up on a completely different trajectory in terms of manufacturing. We won't necessarily go that way, but it's a future that we could face. So, as a small, medium-sized company, I think you have to think about where where you want to go. Are you on a path to be acquired? Uh, is that your trajectory, or do you want to stay one of these rare few uh within the XBI index who are going to be uh independent? And that that defines the trajectory. Uh, but in either case, and some of the choices that you will make along the way, but in either case, I think the key part is understanding what is your value proposition, what is your differentiator, where does that sit? Uh, and I think the temptation a lot of the time is to say, well, everything that we do is brand new, um, and nobody does it like this, and we're the best at every part of uh what we do. But imaginably, this is likely a skewed point of view, right? So, really thinking about what is it that you're eventually selling, let's say, either to the stock market or to the large acquirer, focusing on that. And where do you think others are going to have deeper knowledge, uh, are going to be able to help you better serve you? Uh, and so maybe that's an opportunity to co develop or partner. And again, that's where um the emergence um and the depth now at which AI is in this industry is really changing that landscape. So uh for a small, medium sized company, I think a lot of the We see challenges when it comes to development and manufacturing. So perhaps that's an opportunity to um partner up like we have historically with CDMOs, but now at a data level, partner up with another company so that you can bring in their AI model, let's say, for development or manufacturing. So you don't have to build from scratch. You can already leverage what the industry knows.

SPEAKER_00 22:31

Yeah, thank you, Kat. On that partnering angle, I'm curious to hear more on that, you know, practical advice in terms of partnering. And I mean, when we talk about is it better to build by or collaborate on an internal AI platform as a biotech company? It sounds like you lean towards collaborate. Am I wrong? Or, you know, give me some more color there and we'll move the discussion into that um, you know, build by or collaborate approach.

SPEAKER_01 22:59

Yes, I am uh certainly biased towards collaboration. I think historically this is how we have made it as a species. It's our superpower to collaborate. Um, and um of course, you know, your guardians is gonna say, but what about our IP and so on? So let me just address that in the beginning of this answer. Um prehistoric times, we would have never made it if everybody were, you know, looking for their own cave, building out their own whatever hammers to go after the mammoths or whatever. I mean, I'm not, I'm not, I'm not an anthropologist, so I I can't speak to the details of that technology that they were developing. But we work together kind of in the early stages, and then yeah, every individual from the certain base can then build out their own something that's that's really um uh successful. And I'll quote here uh Bob Young, who um was the CEO of Red Hat, because I think the way that he expressed this was really spot on. So he was talking about software development, um, and how in the beginning it was open source, which is essentially a very collaborative way of developing. Uh, and so the phrasing that he used was that in that open source model, you could stand on the shoulders of others to build whatever you're building. And I think we're undervaluing that approach, the the standing on the shoulders of others, because um, you know, of course, you want your your unique company, you want to build your own IP, uh, you know, and then exit in whatever your um preferred way. Uh but um for new fields, new modalities, new indications, and so on, there's such a value to developing something together to collaborating at that earlier stage. Um and then later on you you know you build out. But of course, to be able to do that, you have to be crystal clear on what is your value proposition, what is the special something that you have to protect, that you have to develop in-house, never show to anybody, and you have to be clear on that early on. Um, so that um again, especially in the way that this is why AI is so um transformative in that sense, because once you share the data, you can't really unshare it. So understanding from the beginning what you want to bring, what's your core strategic value? Um, and then you know, building everything around that, protecting that, but then making the distinction between that and everything else. And so then everything else is fair game for uh co-development collaboration, partnership, uh, right, because you know what your special thing is. Uh, and so in that sense, then the selection of the partner with whom you're going to work um is the next level of strategic decision. So are you going to work with a CDMO? Are you going to work with an instrument provider or you know, technology? I'm going to say more broadly, technology provider. Is it going to be a specialized life science AI technology, or is it going to be big tech? Because they're now entering the industry, right? So you have to for yourself assess uh who is the best partner, who has the best breadth, knowledge, depth, uh, right, uh, for every bit of the workflow uh that you're trying to improve uh with uh with AI. And so then maybe uh there's another level where actually the best partner for you for a collaboration is another therapy developer. Uh and so that's getting to pre-competitive collaboration, right, where you really work on something together, even though superficially that might be perceived as your competitor, but in fact you have some shared goals, some shared interests, uh where developing that knowledge together would be vastly beneficial.

SPEAKER_00 27:01

Yeah, I get what you're saying, Kat. And I agree that you know, as humans, as social creatures, we're we're hardwired to be collaborative. Is there any instance where it would make sense to buy or or build if you weren't uh a huge pharma company that could afford to do so?

SPEAKER_01 27:19

Uh it all goes back um to what is your strategic value? And I think we have to avoid the temptation of becoming software companies. Um, because yes, like many other things, we want to believe that we do that in a very special way, but we have to have some humility and understand that perhaps there's a partner who actually does that better. So um, you know, CDMOs, they will see across customers in great depth uh what the process looks like. They might not have some of the pieces of that data, but but they see many, many, many customers. Uh, and so probably there are some things they really know how to do. Um, then on the technology side, right? Um, I'll give a very basic example. So you can for yourself figure out uh how to detect uh an alert for your freezer warming up, but probably a technology provider would be much better equipped to do that because they have many, many, many freezers and they have seen failures over and over again. They know what the predictive features of that are. Uh for emerging uh specialized life science uh AI companies, um oftentimes they have their own approach to data analysis or their own model that they're bringing. So uh, you know, when a company, therapy developer company makes a decision to collaborate with them, it's probably for that very special model. And then thinking about big tech, they work across customers. So, you know, what we've been seeing so far is that um enterprise-wide deployment of AI is often in partnership with those companies. So again, they've seen failures uh uh across different sectors, across different companies. Um, so there are probably some things that they already know how to handle well. In all of this, of course, uh a question is um how do we protect uh the data and the learning? Um, and that the unfortunate part is that today the only ways that we have of protection are um contractual and social, right? You don't share too much data. And then if it turns out that the partner is somehow using your data in a way that wasn't covered by the contract, you can go after them legally. But these are consequence-based. This is not a technology-based solution. We're all still looking for a technology-based solution to protect the data and the learnings that come out uh of the data. And, you know, with on-prem, it's it's a little easier, but for a company that works across sites, um, on-prem tends to be a very difficult solution. So we're we're still uh and you know, we're still very much developing the technology uh level solution for that. Um, but uh so the contracts have to be very tight. But I think uh just as a as a company, as you're thinking about uh building uh in in the space of AI technologies, remembering that different partners have different capabilities uh and understanding what is your core competence and what is the core competence of each of those types of companies will help make strategic decisions. And then just very practically, of course, there's uh the question of data readiness. Are you actually equipped? If you're thinking about building your own, are you actually equipped to do that? Uh, if you're thinking about collaborating with another company, will they help you out in terms of bringing your data into a shape that it can then uh be learned on? Or uh, you know, are you are you gonna be able to pull that off on your own?

SPEAKER_00 30:58

Uh and so I'm I want to know from from your point of view as a technology advisor and portfolio strategist, what are how are you seeing the industry currently change and evolve that gives you optimism versus what do you wish uh the industry could would do better or focus on? Does that make sense? So like what's the disparity between how we're growing? Like, you know, we're doing this well and we should continue doing that, but um more people should you know there's there's an there's an opportunity here that that is being overlooked at scale. Or I don't know if that makes sense.

SPEAKER_01 31:37

Yeah, I I mean uh again, I think elaboration makes everything much more efficient. So um I wish we were more efficient in the sense of developing faster, for example. Um, and I think an interesting shift there is that discovery with um all of the uh AI technologies that are coming through is almost becoming commoditized. I mean, you could have a new molecule, um, let's say very quickly, um, and not to undervalue the discovery side, but just compare it to how difficult it was historically to where we are today, uh, we're seeing an emergence of just so many options for what you could develop. Um, so in a perfect world, if we were very efficient at evaluating which molecules are worth pursuing, um, and then had depth industry-wide on how to develop them faster, right? I think that that would allow us to get to a space where many, many more patients are served in a much more personalized way. Um because of course, if you have so many different molecules, then now you can really tailor the treatments. Um we won't we won't get into that, but uh, you know, there's an opportunity to really transform also the way that we're thinking about the clinical uh trial space um in this in this world of personalization, where you know, maybe your build is so different from my build that whatever works for you might not work for me, but maybe there's something with an extra carbon or nitrogen on it, and all of a sudden that works really well for me. Um, so um anyway, um your question was about uh where where we should be going. And I and I think that uh for this um collaborative approach, what could really help us? Um, and then you don't you asked in the positive space. So uh with every every bit of optimism, I say this, uh, what could really help us is the increasing financial pressures on the sector. Um, I think um partly uh of course that comes from um just investments going elsewhere because there are other spaces, data centers and so on, uh, where um uh it's been very hot for investments. So I think in some ways we've been underserved um investment-wise by um non-specialists, uh, but also uh just the way that payment systems are transforming globally, um, the way that reimbursements are shifting um here versus uh other countries, I think there's uh a chance that we'll be forced to collaborate, let's put it that way. Uh and uh as um challenging as that might be in the early stages and on the surface, um looking at where other industries have gotten to through financial pressures, um, I think is actually really inspiring. I mean, you look at automotive, for example, which is the efficiency that they've gotten to um and the overall greatness of cars that we have. Um, so I think I think there's a chance that the sector will really, really change. And the financials and AI automation, uh, embodied AI, um, um robotics are all uh pointing in that same direction. So I don't have a crystal ball, but uh I I think uh in a very optimistic uh future, um the there's a very different path that we could be taking as a sector.

SPEAKER_00 35:23

Yeah. Thank you, Cad. Thank you. And uh I do have one last question for you. It's a question I ask all my guests, it's the hallmark question of the show, and that is Kat, big or small, niche or broad, how do you think we can better biopharma?

SPEAKER_01 35:38

I think it's this market level efficiency that could be the first step. Um, right, in going back to the beginning of our conversation, where with clarity around the value proposition of every type of player in this ecosystem as well as every specific player uh within that ecosystem, um, that can allow us to then stay closer to our core lane, where we have core capabilities and core strategic needs. Um, right. And so that that that overlap, once you're clear on what that overlap is, you can then work with others much more efficiently for all of the other uh parts uh of your workflow or your entire system.

SPEAKER_00 36:22

That's great, Kat. Thank you so much. And thank you out there for tuning in to this episode of Better Biopharma, the official podcast of Bioprocess Online. I'm Tyler Manichello, and I will see you next time.