Exploring Artificial Intelligence in Drug Discovery

Hello, everybody, and welcome to Crowncast, the podcast from Crown Bioscience that aims to address emerging trends and viewpoints in oncology, drug discovery, and development. I'm your host for today. My name is Johnny McMichael. I'm Crown's VP of client experience and enablement. If you have joined these podcasts before, you probably heard me speak. You will be used to my Scottish accent by now, and hopefully, babble fish is in your ear. The good news is I've got people on who don't have Scottish accents today. I live in Edinburgh in Scotland. I'm not originally from Edinburgh. It's a bit too bourgeoisie. I'm from a lovely industrial and very scientific town called Dundee where there is, for example, a chap called Brian *** who comes from my home city. You would know him as Logan Roy from Succession. So I will channel my inner Logan Roy on this podcast, but I will do my best to have no potty mouthed expletives. So it's my honor to bring you a conversation with scientific experts from around the globe. I have two of these on today. So on today's episode, this is our first for our Crowncast. This is a *********, our first *********. In actual fact, it's a foursome because Jordan's dog has actually appeared into the room in the last few minutes as well. So today, I'm speaking with two of my tremendous colleagues, Jordan Lim and Kwang Yan. Jordan, firstly, is our associate director of digital product based in San Diego. Jordan has a bachelor's and an MBA from University of California San Diego, Worked in a number of roles, particularly in cancer patient support and advocacy for working at Crown for the last, I believe, seven years. Jordan has worked in numerous roles spanning Crown disciplines. He was a study coordinator and a study director, managed the study coordinators group in San Diego. For the last four years, has been working on the digital side of Crown as a digital product manager and today as associate director of digital product. I will... I'll let... Pass over to Jordan in a moment. And the second guest we have, Kwan Yan, is a... Kwan is a master's and a PhD from the University of Leiden focusing on bioimaging and image analysis. Kwan's actually one of our longer serving members of the team. He he joined Oselo, which is now from Bioscience Netherlands, believe, around about twelve years ago. Juan is, over time, headed up our bioinformatics section there. He's worked in software development with Crown and is now senior principal scientist in our data science and bioinformatics group. I hope I've captured all of that correctly. Oh, thankfully, I can pass over to the guys. So welcome, Jordan, and welcome, Quan. It's an absolute pleasure to have you on. Always nice to speak to you guys. So please feel free to introduce yourselves, your role, and and say hello to the Crowncast listener. Yeah. Thank you so much for having us, Johnny. So, yeah, I think you covered it quite nicely. So I'm associate director of digital product and now operations excellence too. You know, over the last four years, transitioning into digital product, really trying to focus on empowering our teams and both of our... And our customers as well, to leverage digital technology to achieve their goals in more efficient and more capable ways. So it's, yeah, it's a really good role here and, both handling custom software development, as well as digital transformation, here at Cron. Quan is here. I'm a senior principal scientist of data science and bioinformatics department of ChromBio. I'm actually based in The Netherlands. In general, I spent about fifteen years in the machine learning fields, making various kind of platforms for image processing, intelligence. Well, essentially, this is just how I started. But I'm pretty new to Chrome BIOS. I just starting my positions like three years ago. So it's actually fresh to me. But compared to the previous works, I think Chrombolyo is just starting something totally new as new experience for me. I haven't actually done any kind of pharmaceutical related research, but everything is quite interesting, and now we're all starting talking about artificial intelligence. That's that's really exciting. Thanks, Kwan. So for for me, I think the best way to treat me in general, and Jordan and I have worked closely together for several years now, so he knows this well, Treat me as a complete layperson because that's a fairly accurate way to describe me. So tell me, from your perspectives, what do we mean by artificial intelligence? So for me, for example, putting my head onto Patrick Swayze's body doing tai chi in the film Roadhouse is I... Is artificial intelligence. So artificial... Please tell me, yeah, what is artificial intelligence for you guys? And and what is the difference if there is between artificial intelligence and machine learning? So the truth is is that when people talking about artificial intelligence, we're... At this at this very moment, we are actually more talking about, like, a large language models. So it's not it's not really our intelligence yet, but it does have the capacity to understand human languages to do reasoning, depends on the user's input. So... But I think there's quite few people who like to really mix up machine learning with artificial intelligence. So I might want to add some point on this one is that are like intersections. Like a completely two different fields. Lots of people talking about machine learning when they're really talking about artificial intelligence, but other people's talking about artificial intelligence when they really talk about machine learning. So so I do understand the confusion. It's it's totally understandable, really. Yeah. So it's not easy to separate for the for the for the common groups to actually knowing the the real boundary. It's also vague. Yeah. No. I I totally agree. And so AI is a very broad term, basically meaning using machines to mimic human intelligence or behavior. And human intelligence and behavior is so broad, so thus AI is broad. And and so in a classical sense, you know, machine learning is a subset of AI. AI is the broader umbrella term. And some of the, maybe, buzzwords we hear are based around the, I would say, big categories of artificial intelligence that we're seeing come out today. So a few I'll touch on. One is generative AI. This is mimicking a human's ability to create new content. And this is what large language models do, is based on previous learnings, tries to generate something entirely new based on your request. There's also agentic AI is something you may hear, and that is based around decision making and task completion. And I would say machine learning and what is a fundamental to a lot of these is exactly that learning aspect, which is, you know, traditional software is is more static in the sense that you you train it once or you fix rules once. And if you want to update it, you have to do that. You have to update it. But where art... Artificial intelligence brings another level to that is it more so is building a framework where it can continuously learn from new datasets without having to reprogram the application or the tool itself. So it's that learning aspect that is key to machine learning. And actually, on top of that, recently, we actually see, like, deep seats, which is more started focusing on the reasoning aspect of the AIs. So the logical the logical sense and comprehensive abilities of the AI is actually increasing. The reason is that large language model as what we see nowadays is is still not yet enough to comprehend the human knowledge. At one point when when you have sufficient knowledge as as as your as your back end, then the the the intelligent will actually upgrade or progress into more logical and reasonable sense in terms of the in terms of the knowledge processing. So apart from the learning capacities, actually now we are more focusing on reasoning. It's like mathematical reasonings, like summarizations and especially like a logical judgments. Even that's also possible. But for most of the... I think for most of the audience, I think the artificial intelligence is all about ChetGPT and how they see that kind of products. But the truth is, is that if you really look into the potentials of the AI, there are so many different applications that that you can work with the AI. It's actually very flexible, I think. I mean, drug study is one thing and people using these kind of AI engine to predict the stock market trends, you name it. There's a lot of application which is more than capable of doing just a language processing. I think that's actually an important point here. It's not about like having a bot that's giving customer replies or something, but it's way, way more than that kind of point. There are absolutely consumer and business applications for AI, which continue to grow every day. And I guess, well, AI has been inherent in a lot of what we've done in terms of our digital interactions on a daily basis, even using our own mobile phones the last few years. Right? So you finished up there, Kwan mentioning drug development. So could you guys perhaps talk a little on how you see... How, you know, how is AI being used today in drug discovery and development, and how do you see it developing in the coming years? Yeah. You know, it's it's a really exciting time, in in drug discovery with the advancements in AI. I think it goes well beyond what we actually see in our consumer lives. And before I get to specific applications of how it's being used today, I think it's really important to frame, what what the benefits look like in drug discovery. And and I think looking back, Steven Paul and his team, in 2010 published a paper, describing productivity and and drug discovery. And it's essentially a ratio of five variables. On the numerator, right, the positives, you have a volume of work in progress. So the more work in progress you can handle, right, the the more efficient your, your pipeline will be. You also have probability of technical success and then the value both commercially and and, medicinally, of your new drugs. And then on the denominator, what you're trying to balance out is the time and cost to get there. And where we're seeing, the the tangible effects already of AI are being able to boost the amount of work in progress that that companies can handle or generate new ideas around new molecular structures, as well as the probability of technical success, by being able to iterate and help optimize those molecules a little bit further down the line. And, also, with this being now, driven by AI and, less dependency on, you know, human effort to do all of this, it's getting faster and cheaper. And and we see this reflected in companies. And one example is in silico medicine. And silico medicine has 31 ongoing programs, 21 of which have been proposed just in the last four years. You know, 10 of those have received IND approval. And and, actually, they recently published benchmarks on how fast they did this. And comparing to traditional methods, it took them to basically get a new development candidate. It took them between nine and eighteen months compared to traditional ways of, you know, taking about... Let's see here. Twenty four to forty months. Right? So it's it's quite it's quite a long, you know, it it takes the time and half, basically. And how they're doing this, and how how AI is being applied today starts fair at the beginning. So as early as target identification, something that artificial intelligence excels at is combining complex datasets and identifying trends. And so when you're looking at disease pathways and what could be potential targets for drugs, You know, artificial intelligence, people are trying to start combining multiomics data with clinical data, with real world data. And this can produce new insights on how to target these disease pathways. And once you have that, then you can move into generative AI. So generative AI will... You know, the hope here is that we can actually generate new molecular structures to specifically target these targets. And then from there, you get more into machine learning to then optimize those new molecules. So that's where you get into in silico screening or simulations to see if you make any molecular modifications. Will it enhance the binding affinity or some of the properties of your new drug candidate? So, you know, this is how, people are applying today both for small mo... You know, more more so for small molecule research than large molecule, but it's definitely moving towards large molecule as well. And it's... Like I said, the the impacts are already tangible. Yeah. Exactly. And I also... Novel ways to target undrugged, what were considered to be undruggable targets through that modeling. And I really liked the way you portrayed that there as the transition between use of AI to flipping to machine learning. That's probably as good an example to make the differentiation in my head as you could have given. No, I totally agree with what Jordan says. If you look at the workflows, as you're looking to really like into the labs, you start to realize these lab robotics is is basically like helping our stuffs to reducing the times when they spend on the actual pipetteings preparation of the plates. Consider AI is basically the same tools. So it's rather rather than like like helping people with physical labors, it's basically helping you with repetitive, let's say, brain labors. If you look, if look into the if you look what we can actually do with AI, the beauty part is that it's actually a completely black box, but it's capable of learning knowledges and making the relative predictions or conclusion. So you can you can actually fitting in fitting the AI into basically anywhere as far as I can see. I mean, if if let's say in a perfect ideal scenarios, you have robotics connected with your plate analysis and you have the microscope connecting to your plate hotels. And when the microscope is done with the imaging or done with the scanning of the plates, the readouts will be transferred into numbers and the numbers will be immediately interpreted by AI. So basically you're starting with a set of plates before the weekend and you have every conclusion ready by Monday. Nobody has to waste their time doing those kind of repetitive works. As far as I can see, this is what we're actually aiming for. I mean, a sort of like a completely automatons in the labs where you can just initialize your design, put your thought into it, and basically just let the robots and let the artificial intelligence take care of everything. So to narrow senses, of course, you can predict the drug's responses. You analyzing the pattern, the EC 50 responses. But to a top level, is that what what what if AI is capable of reading the numbers and converting the numbers into conclusion and into reasoning and even into a report. So we're basically talking about you have your plate in the machines and then you have to report on your desk by Monday, and then nobody has to interfere. 20 fourseven without kind of, without any kind of interruption. So these are the things that we are actually aiming in terms of pharmaceutical research, or CRO, especially CRO related research. Because I mean, after all, labor cost is one of the main things in the process of of the research. So I would say this is this is way more than just a cost efficiency. This is like fricking people from these tedious works as far as I can see. Some colleague told me that they spend like two, three days of preparing reports just for the customers. AI do this in two hour. So you can do the math math. You can do the counting. That's the efficiency that we're talking about. And what is nice is that AI don't need rest. Okay. Yeah. Fine. You can force it to work. AI doesn't need rest. AI doesn't have emotional issues. This is why you can just trust it with the repetitive works. I mean, of course, I mean, trust to a certain levels. But that's what I mean. I mean, imagine you have to, like, we spend like not spend, but we're asking our Ph. D. Staff or even master staff to actually sit in front of the computer for three days just to writing a report, making figures, making plots. I mean, that's... I've been there. That's why I... So are the confidence levels high? Because I mean, that's... I love your angles there. But as it stands now, and a lot of these processes that you were taught there, Kwan, we still need that human element. Right? That QC or... Right? We're probably not at a stage where we can eliminate that. Right? That I have to say is one of the main gray area of AI, really. The quality controls, the validation of the results. That's actually the most difficult topic that we have at this very moment. You can consider AI as a person, very sophisticated person, which don't need any kind of rest. It doesn't have emotional issues. But the problem is that as AI, it's basically essentially a human mind. It does make mistakes. Knowledge. Its capacity. So the AI's capacity is largely relies on the qualities of the data or quality of the knowledge it has been fed with. So if we're talking about inconsistencies or errors or even contradictions in the data itself, then like every single human, it makes mistakes. That's why you can't really completely... Let's say you just can't completely left out the human factors when working with AI. And I never agree with anyone saying that, oh, AI is gonna replace human. That's not gonna happen. Not really. Not until the point that AI becomes completely sentinel, not like the Skynet or something. But to that point, you really have to put a validation process. That's always the case. You can't skip that. That's impossible. Like, for example, we work with some of the data and what we realize is that on average, AI made roughly about 10 to 20% random choices, which is not there. So maybe it's just our particular data sets. But when it comes to like a numbers or these kind of things, AI intend to create artificial result, which is not in the original data. So we actually paid very close attention to these kind of problems in past few months, not actually months, but actually years to actually try to fix these kind of problems. No, no, no. I would definitely not gonna say that you can just take take out of the human factor. That's that's not gonna happen. Not not not in any chances. Yeah. Just to to follow on that, you know, I I... You know, a key concept is human in the loop for all of this. And when I... When when someone's trying to use AI for the first time or even in complex scenarios for drug discovery, the way I frame it is how can AI do the first 70% for you? It... It's really good at getting you started. And I think that this is actually the beauty at where drug discovery is using artificial intelligence already in a meaningful way today, is we're not, you know, bringing a drug to market just because it was proposed by artificial intelligence. Instead, we are doing things like target identification, generating new molecules, but all of these need validation. They need a first pass by a human, and they also need experimental validation. And so we're not actually relying on any decisions. There's no decision making by artificial intelligence. What it does is it lightens the the burden of ramping up a task or doing the first the first bit so that, you know, very skilled individuals can focus on what is most meaningful, you know, the the most meaningful use of their skills. And also, it it unlocks the ability to apply some of your effort towards those creative tasks, which is is aspect in any kind of job. So that's that's kind of how I I look at it today. So can we trust it? I'd say as as something to do an initial proposal and to get you started. I think it it can be quite good. But for a finished product, definitely needs human in the loop. Thanks, guys. That's, yeah, really, really fascinating. I really love hearing your thoughts on this. And it's probably... These are the kind of concerns I think that the folk out there will, you know, will have, well, particularly those not working with AI. Those who are working in AI are probably already fairly convinced. But... So in terms of drug development, drug discovery and development, are there any are there any regulatory hurdles that need to be surmounted for use of AI? There's foundational regulation around things like data privacy that will continue to apply and maybe even be more important given how data driven, some applications of artificial intelligence can be. So looking at things like HIPAA and GDPR, you know, protecting the privacy, especially as you start to loop in real world and clinical data, absolutely need to be adhered to to support data privacy. And then moving more specific to artificial intelligence, there is legislation that is starting to come out, like the EU AI Act, which is a risk based framework for regular, regulating, AI technologies, trying to categorize it based on risk. And the FDA and CEDAR, so the the Center for Drug Evalu... Evaluation and Research, has taken a similar stance. They've essentially said, AI will undoubtedly play a critical role in drug discovery. It's it's inevitable, but, we want to do it safely and efficiently in a way that benefits patients. And so they too are taking a risk based approach to regulating AI. But I think that there's a general sense of collaboration that people want to move in the direction of innovation. We just need to do it safely. And and the last piece I'll mention is something that, yeah, has to be has to be resolved is intellectual property concerns. Because with artificial intelligence, a lot of these models are being trained on potentially vast datasets and potentially datasets people may not own. Maybe publicly available, but they don't necessarily own it. And so then when you get to things like generative AI, which are generating new outputs or coming to new conclusions, based on this data, there's a question of who owns the output. So there's no good answer today. But these are some of the things that we need to be mindful of when trying to leverage the technology. Mean, I to the technical front, I mean, we always talking, I mean, in terms of like machine learning engineers or artificial intelligence engineers, we're always talking about this topics is called making a responsible AI. So this basically is is to make AI in to fit human morale standards or or to to to make AI to actually fit into legal boundaries. Reason is is once you actually feed your data into the AIs, even if you even if you try to put a restriction on the AIs responses later, there is always a way or there's always a logical tricks where you can basically tricks the AI to give you the answers, even if it is not allowed. So that actually leaves a lot of holes rather than just just the the the the yeah, of course. I mean, like a standard data protection, like GDPR's. Traditional ways we know where the data is. The only way you can get the data is hacking the server. But for AI, we are actually talking about a set of knowledges where you basically taught AI to learn. And the later you tell the AI saying, oh, you cannot tell this information to the others. How can you actually do that? You can't. I mean, it's it's a logical paradox. Has the knowledge. Okay. It has the knowledge and you tell it to actually stop giving that knowledge to the people. There's always there's always a bit of like a way that you can always bypass that kind of restriction. Very, very difficult to be frankly. I mean, tried I tried different kinds of engines. I mean, they all failed. It doesn't matter. It makes no differences. But that's that's the issue even with our own ChromBiles data at this moment. I mean, we are actually facing these kind of dilemmas as well. We want the AI to provide the knowledge, but we don't want the AI to provide particular knowledge to particular user groups. Then, yeah, it's difficult. Unless... So the only solution is don't tell the AI. So far, what what I can see is just if you don't want to have the information in the AI or you don't want the AI to tell the information, just don't even bother to give it. That's probably the most safest way. But other than that, I seriously doubt. Maybe it's a little bit pessimistic, but I've seen a lot of loopholes inside AIs when when we're talking about really confidentiality or intellectual properties. It's rather than a gray area. It's it's already something that's already literally happening at this moment. So it's really a pressing matter that we need to resolve as soon as possible. I don't know. I mean, personally, I don't have a solution yet. I really want to hear from other specialists to see how to... It's it's it's it's really it's really more than just a programming problems, really. Some of the listeners here actually might feel comfortable enough to reach out to you at some point, and then we discuss some of these topics directly with you, actually. That would be a nice thing. And I think that's where, you know, you you marry the technical side with the the regulatory frameworks and the and the legislation that's coming out, which is similar to my product side. You know, trying to set the requirements as to what needs to be met and then working backwards into the technical solutions. Because it can be it can be like you're trying to boil the ocean, especially with this new technology trying to fix everything. And so, you know, things like with the EU AI Act, some of the things that it categorizes as the highest level of risk is things like social scoring, which takes into account, let's say demographics and economic status and things like that. So social scoring, biometric identification, these types of things that may invade privacy from a functional standpoint, then trying to see how can we utilize the technology still in a meaningful way while meeting these requirements. It's a bit pity or not pity, but I think now everybody is... Believe that the benefits that bring abroad by the AI is outweighed the potential risk. Well, I personally have to agree that indeed, at this very moment, it is the case. We definitely can take a risk. Just have to be very careful. But there is a Chinese saying that if you got choked once, should you give up eating? Probably not. So we actually learn by actually doing. It's a very powerful tool, just like a web. The first time when human understand the genomics, same things. I mean, maybe at one point we'll start to create dinosaurs just like Jurassic Parks. The same thing goes for AIs. I mean, yeah, it can be a very helpful tools or a very helpful friend even, but it can be also something that complete breakdown your life. But I prefer to look into the bright side. Let's try that nevertheless. So do you guys have any more... I've got... I pretty much just got one last question to throw at you. Do you guys have any further comments on the applications of AI and drug discovery? Yeah. I mean, I'd I'd maybe drop in one more one more note. You asked earlier, where do we see this going? And I think on the the grand scheme level, the the direction that we're headed with artificial intelligence is... The aspiration is to move from drug discovery to drug engineering. And what I mean by that, if I use an analogy, today, there are lots of locks associated with diseases. And we try to find a drug or a key that fits that lock. It's a very difficult process. And it is discovery in the sense that, hopefully, you you may be... Let... Let's take, like, large molecules, for example. You may take millions or billions, toss it into a a beaker, and see if one of them hits your target. This is a discovery process. It's very experimental. With artificial intelligence, we're moving from finding the key to designing the key, to fit the lock. It's it's a very challenging problem. It's much simpler when it's a key. But that is the overall direction. And and how this technology empowers drug discovery is to have those design and iterative capabilities to, yeah, do things that were otherwise impossible or nearly impossible Very nice. Yeah. That's a lovely analogy, actually. It's really visual. So thanks thanks, Jordan. Good. Good. So, obviously, today, we're we're we're not here to to talk about Crown Bioscience as such. So I did just want to ask without giving any secrets away, of course, if you could give us a few little insights into the ways in which we, you know, use AI currently within Crown as a as a as a CRO, any, you know, any initiatives that we're working on involving the use of AI? Yeah. So so I have the pleasure of working very closely with Quan. We actually work on the AI program together here at Crown. And so what I'd say is, generally, our goal is, you know, as Crown, we're an insight generation company. And similarly, we're trying to leverage artificial intelligence to unlock new insights for our customers, to guide key decisions and actions, throughout their engagement with us. And so, those include things like, you know, what... Which cancer model is going to be the best for my research? Or how do I optimize my study design, you know, on the front end? And then on the back end, how can I pull the most insights from the data that we've produced to then plug into the next steps? So this is where we're trying to leverage artificial intelligence is to really, in a sense, provide an advisory role for our customers to make more informed decisions. We've been working with, like, couple of... Actually, nearly all of the Crown Bio sites. And the the main focus of our AI team is to actually offering Indeed consultant and adversary knowledge using AI systems to actually make like experimental design more efficient. Indeed, that's actually the direction that we want to go and make the like. Think also it department is working on making our knowledge management or even business intelligence become more efficient in terms of I'm not going to say number processing, but I'm going to say knowledge processing and then reasoning and etcetera. So there's there's actually a lot of things you can do. I mean, if you see if you see our project list, there's like like a two dozens of projects related to artificial intelligence. And and and at one point, yeah, but at one point there's going to be like a boomings of all kinds of supporting mechanisms that we can introduce to our daily works. That's going to save people a lot of troubles. I mean, are showing people the prototypes and people just love it. And the beauty is that once we set up the systems, we just have to feed the AI with more knowledge. That's it. There's no additional programming. There's no additional, let's say, workload that really on top of that. But setting up the framework is not easy. So in that sense is where our job is mostly focused on that aspect. But at one point, let's say we would call it a threshold. If you're crossing the complexity of threshold, then AI becomes really sophisticated of all kinds of questions, especially not not yet in terms of like a really like a providing designing, but optimization. I mean, AI is... To my surprising, AI is really good at optimizations. So if you give them the conditions and knowledges, it'll work out much better solutions than any human can possible comprehend. So that's that's something that is that we can take a huge advantage of. And this is also something that is really valuable for CRO company like us. So although I won't be able to talk too much on top of it, but I'm really, really exciting to see what happens in the next half years And we'll see a lot of things change and it completely tipping down the lab lives, let's say, of most of our colleagues. That's going to be really exciting to see, actually. That's why Jordan and I were actually to work closely on observing the changes and introducing the modifications and etcetera. And it's a learning from feedbacks from our potential users. But those kind of things are only for internal, not yet for the public, because we haven't resolved the data protection issues yet. Once that point is resolved, then yeah, there will be some tools released to the customer. But at this very moment, data security, some confidentiality is our major concern. Mean, you have no idea how many times we will raise up this issue during the discussion. Oh, how can we protect the customer's confidentiality? Okay, I will do that. But the truth is is that it's it's more difficult than it looks like. Obviously, a key focus for any any CRO is to to ensure those those concerns are all taken care of upfront. So any closing thoughts? My end, yeah, really appreciate you hosting here and inviting us, Johnny. It's been fun to to talk about this and and share the time with Kwan here. Thank you very much for the choices. I only have one take home messages for the year, maybe for our potential audience is that being a black box like an AIP and black box is not actually a bad things. So what what you can consider using AI is that it is capable of building up all kinds of relationships that you can imagine. So if you have an input, if have an output, then the AI can just fill up the gaps. So if you want to use AI, of course, first easy thing is come to us and ask for us to do it. But on the other hand is is that just put touch your imagination to it. If you want to describe a relationship, put an AI in there, put the inputs and put the output. That's it. Don't bother to think too much. Right? I mean, that's the beauty about the AI part. So that's the message I would like to really deliver. I mean, essentially on top of all kinds of AI applications. Yeah. I would say that people would start to realizing they're thinking too much on top of AI. They can use them easily. No worries. But easy ways, of course, come to us. Maybe a brief call to action, you know, give it a try if you haven't. Whether it's your personal life, professional life, in a safe manner, definitely give some sort of tool a try because upskilling in this area is going to be a key to how, you know, the the industry roles are shaped in the future. Thanks, Jordan. Thanks, Quan. And to those out there listening, I hope you've also enjoyed the discussion. Thank you for tuning in to Primecast. Please do like or subscribe on YouTube or Spotify. You'll also find us on... Directly on the the Crown Bioscience website, crownbio.com, We hope to have you join us again for a future episode of Crimecast. Thanks, guys. Bye bye.