New Matter: Inside the Minds of SLAS Scientists

Lab of the Future | The Integration of AI/ML in Life Sciences Labs with Trevor Vieweg, M.S. (Sponsored by Molecular Devices)

April 25, 2023 SLAS / Trevor Vieweg Episode 148
New Matter: Inside the Minds of SLAS Scientists
Lab of the Future | The Integration of AI/ML in Life Sciences Labs with Trevor Vieweg, M.S. (Sponsored by Molecular Devices)
Show Notes Transcript

Continuing with our "Lab of the Future" series, we welcome Limmi CEO and Co-Founder Trevor Vieweg, M.S., to the podcast to discuss how artificial intelligence (AI) and machine learning (ML) are integrated into life sciences labs. Vieweg shares his perspective on the lab of the future and Limmi's approach to integrating AI and ML, and how AI/ML can revolutionize the life sciences industry.   

For a transcript of this episode, please visit this episode's page on Buzzsprout.

Key Learning Points: 

  • How integrating AI/ML differs in life sciences labs differs from the tech industry
  • Solutions to the biggest challenges researchers face with AI/ML
  • The future outlook on laboratory hardware and software advancements
  • What AI/ML integration looks like for small to large-scale companies

Our Sponsor for this Episode
Molecular Devices makes scientific breakthroughs possible for academia, biopharma and government customers. Dedicated to enabling life science labs of the future, where innovative technology and novel research meet, Molecular Devices empowers scientists to advance discovery, driving earlier diagnoses and safer therapeutics for patients. Spanning cell line development, 3D biology and drug screening, their automated, end-to-end solutions streamline and scale complex workflows while integrated machine learning-enabled analytics allow researchers to mine data easily for insights.

Molecular Devices is the innovation partner that empowers scientists with next-generation technology to advance discoveries, improving the quality of life everywhere.

Learn more about Molecular Devices by visiting:
https://www.moleculardevices.com/

Stay connected with SLAS

Upcoming SLAS Events:

SLAS Building Biology in 3D Symposium

  • April 16-17, 2024
  • Jupiter, FL, USA

SLAS Europe 2024 Conference and Exhibition

  • May 27-29, 2024
  • Barcelona, Spain

View the full events calendar

Hannah Rosen: Hello everyone and welcome to New Matter, the SLAS podcast where we interview life science luminaries. I'm your host, Hannah Rosen, and today we'll be continuing our series focusing on the lab of the future with Trevor Vieweg of Limmi. Limmi was one of our SLAS 2023 Lab of the Future companies, and Trevor is joining us today to discuss the integration of AI and ML into life science labs. Welcome to the podcast, Trevor. 

Trevor Vieweg: Thanks for having me, Hannah. 

Hannah Rosen: Our pleasure. Did I pronounce your last name correctly? 

Trevor Vieweg: Yeah. 

Hannah Rosen: OK, good. I forgot to ask beforehand. All right. Well, we're happy to have you here today, Trevor. I was hoping that you can maybe start us off by providing us with a little bit of your professional background. 

Trevor Vieweg: Yeah, sure. So I've spent the last 15 years in systems engineering, data and software roles across primarily defense, aerospace and consumer markets. So I've jumped around quite a bit in different industries, but spend a lot of my time in highly regulated industries bringing new technology and the data and software race to those industries. I'm one who stays a little bit restless, so I like being right at the intersection of technology and product and business. So I basically flip flop every few years, leading engineering teams or more business focused roles. We started Limmi, my co-founder and I Brian, after we got involved with a local pharmaceutical company here and it was really our first venture into the biotech space, but we saw a huge need for better tooling in the AI and ML space. Also, just a massive potential across the industry to start to use these tools. As you know, the AI space, as probably most of us have heard, has really matured and so that's how we found ourselves in the life sciences space. 

Hannah Rosen: Wow, quite a journey. That's... that's fantastic though because I imagine you probably have a much different perspective on some of these things than people that perhaps have been in life sciences their entire career. 

Trevor Vieweg: Yeah, we... we do get some comments about, what are a couple of aerospace engineers doing in the life sciences industry now. But I think it really helps to be honest. There's enough overlap with some of the, again, you know, dealing with the highly regulated... regulated space that we have had a lot of experience in. So we're very sensitive to that. Obviously, a lot of that technology can be used in the life sciences space as well. You know, we think bringing a little bit of a fresh perspective actually helps with a lot of these conversations, so long as you're really deeply listening to folks that are very entrenched in the life sciences space and understanding a lot of the nuances that that come into. 

Hannah Rosen: So speaking of perspective, you know, how do you guys at Limmi view this idea of the lab of the future? 

Trevor Vieweg: So when I think of lab of the future, I think of a lab that is much more software and data-driven than... than systems are today. So really I think the... the big revolution in labs over the past, maybe 20 years, has been around automation and, you know, starting to bring more hardware, automation, or processing into the lab base and that's generated a huge amount of data that a lot of companies have captured. But as we talked to customers in the life sciences space, people aren't able to tap into that data and really harness it all that well. And so I see a lot of new innovations coming in, being able to access that data and making it much more accessible to researchers so they can use it effectively and then using AI and ML to help basically across the entire life cycle of... of life sciences product development to really transform a lot of those areas. I think we've seen similar things in other industries where, you know, you had a hardware revolution followed by a huge data and software revolution. And I think life sciences is really in the middle of that. So when I think of lab of the future, my hope is that we're driving towards a world where software really is... is not a pain point for researchers anymore, but really a tool that they can harness. 

Hannah Rosen: Yeah, that's really interesting cause it does make me think, you know, a lot of the times we compare where automation is in life sciences to some of these other industries that are further along, but you bring up something that I hadn't even really thought about that much where, you know, if we look at maybe some of the other industries where automation has really taken off, like manufacturing for, let's say you know, the automotive industry that's one of the go tos of automation that is just fully implemented, but they don't deal with this massive amounts of data right, as much. Or maybe I'm... I'm, you know, off base since I haven't worked in that industry very much. But it seems like there... it's a lot more of just, you know, you're assembling something and off it goes, whereas with life sciences, you know, we're using this automation and our end goal is these massive amounts of data, do you feel like that has an impact on how automation can kind of be integrated into the industry? 

Trevor Vieweg: 100%. I think that, you know, an interesting... just like you mentioned, right, automotive, you know, oil and gas. I bring up again, it's a very simple product, right? But obviously the reason I bring it up is it's a very valuable product, right, to at least the world as it stands today. But these are simple products, right? And so they've... they've incorporated a huge degree of automation in their processing. But they're relatively simple from the standpoint of, you know, you get your manufacturing process down and away you go. I think the interesting part in life sciences is that automation could be used in the research area in a way that can't be used in some of those other industries. And so you do get this huge amount of data off of what ultimately is a very complex ecosystem, right? Life science is, almost by definition, is dealing with very complex biological models, even in its most simple use cases compared to those other industries, and that again makes it a great target to do more data science work and bringing AI and ML out of the space to help researchers because it's a very tough problem for any one human or team of humans to sift through all of that data. 

Hannah Rosen: So in your experience, you know, where do you see and how do you see the life science labs currently using artificial intelligence and machine learning in their lab spaces? 

Trevor Vieweg: So as a general rule, I would say teams are not using much AI/ML right now and that's obviously something that... that we think is important for them to start using and we want to change. But I... I think what we have seen is some... some really great, I'll say early-stage companies and early potential for what the... what the industry can look like. So you see a number of companies coming out with a... with an AI first or computational model first approach to drug discovery where they are, you know, not using traditional methods of drug discovery, but actually using simulation and AI to help them discover new drugs right from the get go. You see other companies that are specializing in new target discovery using AI/ML. So again, the human genome, huge amount of data, a huge amount... amount of possibilities to sift through for different targets there. That's a great space to be using AI/ML and you've seen some... some really promising results come out of these companies. That I think will become just the de facto way to do a lot of this work in the next few years. One other area that, you know, I think is a standout is around personalized medicine. So I am definitely a believer that personalized medicine is the... the future for us. I think we've seen some really just incredible results around cancer eradication. A number of different areas, particularly around personalized medicine, that's a huge amount of data if you think about you as a person and the amount of data that you can collect about your health, side effects of drugs, how you're living your life, lifestyle choices, right? These are all factors in this personalized medicine approach. Again, this is a space that AI and ML can help, and you see some companies doing that really right now, with some really promising results. So my hope is that we'll see all of those trends continue and then expand into, you know, some other areas around manufacturing and around just general preventative medicine. I think there's big opportunities there as well. 

Hannah Rosen: You have, obviously, a very strong background in many different industries that are utilizing automation, AI and ML, which I imagine is a huge advantage cause you can kind of come at the life sciences with this perspective of, you know, where the gaps might be that we're not aware of in this industry. So, you know, in your opinion, how does the use and mentality surrounding AI and ML in the life sciences kind of differ from what you'd see in tech, or even just other industries using this technology? 

Trevor Vieweg: So I think the mentality around AI and ML in the biotech industry has been... has been somewhat skeptical, but also, you know, very curious about the enormous potential there. And it doesn't look too dissimilar from what the tech industry looked like, I'd say 7 to 10 years ago, which is there was a lot of promise and articles about how AI could transform the industry. But a lot of folks were not sure how to use it and very skeptical about, you know, what problems they could solve. And we see a lot of the same mentality, I think, in life sciences as well. Some of that skepticism is honestly warranted, right? Life sciences, as we were talking about, is a very complex industry. Almost all the problems that are being solved to deal with very complex biological systems. And so there... there is a fair critique. This technology can't be applied just like it was in, you know, the tech space, particularly the consumer tech space, because you're dealing with a more complex problem, you know, solving personalized medicine for cancer eradication is a very different problem than optimizing your social media feed by looking at, you know, what other your users are using, right? And, you know, the other piece of that is because it is human safety. You need to make sure that it works every time. There is a very low tolerance for failure or for bad outcomes coming out of that AI. And so I think justifiably to some extent, the life sciences industry has moved slower than this space. The difference that I see now, particularly as we... as we talk to folks is the... the maturity of AI and ML systems has really, you know, as we've all read about gone up by, you know, many... many... many factors in the last 10 years and so that... that maturity of that technology really at the point now where you can start to use it across life sciences. 

Hannah Rosen: Yeah, I was just wondering cause this is something that I have noticed recently is, you know, with technologies like, you know, ChatGPT, everybody's talking about it, not just in the media, but it seems like almost every, you know, biotech person, everyone that I talked to in the past couple of months is bringing up ChatGPT and how it is changing everyone's perceptions of what AI can do. Is this something that you're seeing when you talk to your customers? Has, like, has there been a shift since ChatGPT has rolled out in people maybe being more accepting of AI in the lab spaces? Or is it causing people to become more cautious potentially? 

Trevor Vieweg: So it's kind of an interesting effect that ChatGPT has had, I think like a lot of AI, which is it becomes a little bit of a mirror for what you believe about AI in general. So some folks we talked to see the results of ChatGPT and are incredibly excited and that's great. We... we think that's actually the... the right way to view ChatGPT. It's going to be an incredible tool that can... that can help across the industry as well as other AI models that are out there. Some folks read about, you know, you've probably seen articles about the hallucinations as they're called by ChatGBT or the fact that it isn't always reliably delivering information, you know, have some, again, very genuine concerns around that. I think the part that's very hard for anyone to visualize is just how fast this technology is going to get better. We actually have had some conversations with customers who are seeing that between ChatGPT 3 and ChatGPT 4. If folks have been using that, they see the difference in how much better this... the system is getting, and I think that brings them a lot more encouragement to say, OK, the time is right for us to use AI and ML and start to bring this into our environment and operations. 

Hannah Rosen: I wonder, you know, do people express concerns because this is something, I think, that I would be most concerned with in bringing AI and ML into my research space, is just making sure that because, you know, a lot of the ways with all of this data, we're really trusting the AI and the machine learning models to know what data is important and what data we should be looking at and what data to pull for our analysis just because there's so much data to sort through. What risks are we looking at for trusting the AI and the ML to be pulling what's important and, you know, how concerned should we be that we can, especially with how quickly this technology is moving that we become over reliant on just believing that the models are... are looking at what, You know, we really should be focusing on? What's important. 

Trevor Vieweg: So you touched on a couple of the main, I think, challenges for AI and ML as we move forward, which is around, you know, one of the fundamental problems that across every industry is what's known as the black box problem, which is you depending on the AI model that you're using. This is particularly a problem in machine learning based systems. You can't necessarily directly see how that system is making the decision that they are or the recommendation that... that... that... that it is. You know, there's a lot of research going into how to solve that problem. There are actually some techniques that exist today that can mitigate a lot of those concerns. Upfront, but it is something that... that... that you need to watch for. You know, I think life sciences could really take a lot of lessons from the technology industries in how they roll out AI and ML. So as you mentioned, another big concept is around monitoring the AI and ML system. Once it's in a production environment, making sure that those outputs that you're seeing are consistent with, you know, the goals that you set and what you're trying to achieve with that model and that, you know, it's not, as is called in the industry, drifting over time, right? So model drift is something that you have to watch for. Again, the life sciences has benefited from waiting a little bit here because a lot of these are problems that are being solved actively in the tech industry already. And so life sciences can really benefit from that. 

 The other area that you mentioned that I think is important for people getting started at AI and ML to understand is around that... that data set and what you choose to feed into your model. And so AI/ML, they are capable of handling huge amounts of data and processing that data, but they're not yet, you know, an artificial general intelligence. They're not going to be able to decide what data is important and what data is not without some training first. And so what we talked to a lot of customers about in this space right now is AI is a tool. It's a very powerful tool, but if you don't know what type of outcomes you're trying to get out of... of the system and where to start looking for it in your data, you don’t have to know exactly, but you need to know a little bit to start to help the... the model understand what type of factors it should be looking for. That's the first thing that we need to go align on with the customer for two reasons. One, you're not going to get a good model if you don't go do that. The old adage, garbage in, garbage out, definitely applies with AI and ML. But the... the second piece of it is we shouldn't be trying to make AI and ML a system that we're so reliant on that we're not solving problems as humans as well. I... I definitely believe in AI and ML being a tool that can be leveraged by humans and then humans making that ultimate decision down the road rather than, you know, it being something that takes over for these humans. 

Hannah Rosen: You know, I... I know I have some friends who work in the tech industry and, you know, have experience with some of these machine learning models and... and with AI. And I know, you know, there's been a lot of lessons learned in that industry sort of, kind of, what you were alluding to with, you know, putting in putting the data and... and inputting our own human biases in with that data as well, and having that be reflected in those models then. And so I was wondering if you could go in a little bit more about some of the lessons learned in other industries when using AI and ML, and how those lessons learned can be applied to the life sciences industry. 

Trevor Vieweg: You touched on a couple of them. So I think there's been a lot of lessons learned in how to think about training and validating an AI model and you'll see a lot of discussion around things like overfitting or underfitting a model, basically meaning, you know, is that, is the model being too strict or too loose with, you know, the factors that it's considering to make a decision on the... on the back end. I won't go into too much detail there, but I think that the overall lesson learned is much more thoughtful about how you're training a system is... is important, particularly when you have a complex problem where you're probably trying to solve one particular area like life sciences. The other piece is around that monitoring side of things. So in the early days of AI and ML people, I think, assumed that these models would behave the way a traditional software module might behave in production, which is, you've done testing, it's static, it's going to keep doing exactly what you think it should be doing once you roll it out into production and quickly discovered that, you know, the real world is messy, which is part of the reason that AI and ML is so powerful.  

But you will get new pieces of data fit into your model once you get into the real world that you weren't expecting, and you need to make sure that that's not affecting the model's ability to produce quality results. Previously on, I'm blanking on the year or the event, but there's been some famous financial cases with stock market investing where we've seen the impact of that, right, and seeing, you know, massive drops in the stock market because people are using these types of algorithms or AI models to drive their invest. And suddenly you see something in the real world that you never accounted for in your training data and get a very bad result out of it. And so I think all of these are things that the life sciences industry can take advantage of from folks. And it's... it's a really exciting pairing, to be honest, Each time we talk to customers, and I know other companies are as well, which is bringing that knowledge and lessons learned from the technology where we can help, you know, the life sciences industry not trip over certain areas that the tech industry definitely did. And then, you know, learning about the specific problems or... or interesting areas of biology and learning some ways to consider a model that, you know, we might not have previously just given our tech backgrounds. It's a... I think it's a great partnership between the two industries. 

Hannah Rosen: Yeah, absolutely. You know, you're coming in it with your perspective on things from the tech side and then you've got life sciences scientists coming in with their perspectives. And it just seems like a perfect pairing where, you know, you're gonna be thinking of things they haven't thought of and they're gonna think of things that you haven't thought. 

Trevor Vieweg: Yep, yep, absolutely. 

Hannah Rosen: Yeah. So what... what you're saying there, it kind of to me really emphasizes that point that I think a lot of times we will make where, you know, every time we talk about AI and ML, people get concerned of, oh, it's gonna replace all these jobs. You know, I've seen all these articles where it's like, you know, 50% of jobs in this, you know, sector are gonna be obsolete by 2025 or, you know, whatever they like to say. But what you're saying sounds to me more emphasizing the concept of it's not replacing jobs, it's changing jobs. So how do you see, you know, the implementation of AI and ML changing the jobs in the biotech field? 

Trevor Vieweg: You nailed it, which is, it's... it's going to change a huge amount of jobs. I don't see it replacing a huge amount of jobs. And I think that generally it's going to change these jobs for the better. So just kind of a... a fun back story. You know, a lot of, I think, software folks are responsible for the type of replacing jobs, but for... for those of us that are a little bit older, read some of the history, actually there was the same argument around a different technology breakthrough many, many years ago, which was the... the keyboard and mouse and some of the first user interfaces for a computer. Some folks said, you know, were the... the death of software programming jobs, because now everything was going to get so easy that everybody could do it, and a software developer wouldn't be needed. So I think you see this every time technology comes out for the most part. My belief is, it almost always is a tool that makes people's jobs easier, better, and allows them to solve more complex problems.  

I think life sciences is probably the industry where that's most true. We are consistently blocked in the life sciences industry by the complexity of interactions of, you know, different pieces of the human body when talking about drugs or... or new discovery for therapeutics. It's just an overwhelmingly complex system. AI and ML are just going to help us get to the next level of being able to deal with some of that complexity overall. So in terms of how jobs will change, I think in the next few years that the overwhelming positive will be that bioinformaticians, data scientists, researchers will spend a lot less time trying to figure out how to wrangle their data and get it into one place. I think that you see a lot of companies, ourselves included, are helping with, you know, just getting that data into a format that's easily accessible and could be used both by AI models, but also by humans. And then the next big space will be not spending as much time doing manual analysis, or what I'll call kind of a tedious analysis side of of... of work. So AI and ML can crunch through those giant data sets that researchers are having to crunch through today. They can replace your, you know, 10 million line CSV file with just a model that spits out results for you, and that's going to be some of the low hanging fruit that researchers, I think, will be very happy to get. A little longer term, I think AI, and you see kind of the first start to this with cheat GPT or some of the ChatGPT, I think that AI can serve as basically a research partner, so can serve as a, you know, a... a colleague that you can ask questions too that, you know, is really good at sifting through large amounts of data. But then, you know, needs the researcher to point new areas to think about or some of that creativity and discovery and synthesizing kind of high level strategy and problem solving. So, I think it's going to be a really great transformation for pretty much everybody in the life sciences space. I... I don't worry too much about jobs being eliminated because I think even jobs that are eliminated, new job always pops up, you know, because of those transformations, that's probably a better job than the last one. 

Hannah Rosen: Yeah, I love that perspective of AI being your recent new research partner, I think that makes it sound a little bit friendlier than... than how a lot of people will tend to think about it. 

Trevor Vieweg: I think, you know, the... the interesting thing is, for all the skepticism around AI, there's also, I think we're all a little guilty of it. And maybe it's, you know, movies of, you know, you hear the word AI/ML and you think of this kind of all knowing robot that, you know, in many scenarios is, you know, kind of bent on destroying human civilization or something like that. But I think we're all a little guilty of when we hear the word AI/ML thinking that it's going to be able to solve every problem and, you know, just not need humans. And I think we're a good ways off from that still. I think that it's going to be much more of a partner than anything else. 

Hannah Rosen: Yeah, I really like that. You know, so, we've talked a little bit about, you know, the concerns around the black box that is AI and machine learning. Are there any other really big complaints or problems that you hear from... from life science researchers who are trying to use AI or ML in their labs? And in those cases, you know, what solutions are you able to offer them? 

Trevor Vieweg: Yeah, the... the other two big ones that we hear are... are... are... what we're really focused on and how we started the company. So the first is around accessing the data. Every analysis problem starts as a data problem. If you can't get access to the data, particularly in... in AI and ML, you're never going to get a model off the ground because you need to be able to train it. A lot of companies have a problem that you know could probably be summarized as they have a huge amount of data. It sits in 20 different silos and you know is in different formats and they don't know how to access that in a in a centralized space. And so we help companies with our platform aggregate that data. Put it into a a format that can be used by an AI model, but also the side effect is, but if they're not using AI yet, it's just much easier to access that data as a researcher as well. There's a... there's a lot of great work going on by a number of companies to do that, not just within a single company, but across companies as well. And I think the more that the life sciences space can figure out how to share data across different groups, entities, companies, obviously some of that is proprietary or, you know, something that a company doesn't want to share. But the more the life sciences can do that, the better the outcomes for human health are going to be inside. I think a lot of good work is going into that space by a number of companies. The other big challenge is just the difficulty of the, I'll call it the ML too lean ecosystem. So if you want to stand up a machine learning system at your company, you have to go wrangle a dozen or so different software tools to get your data pipeline set up properly, build out your training data sets, evaluate how your model is doing, deploy that model, and that's a lot of work for folks in the life sciences space usually requires a number of software engineers to be in and helping them. And so what we're really focused on at... at Limmi is making that something better. Researcher can use out-of-the-box without having to have a lot of software support for it. Once we're able to do that, that's really, you know, the way that we see this scaling out across the industry is... is if you remove some of those barriers to entry for researchers, give them the tools to go solve the problems that they want to solve. That's where we can unlock, you know, using AI and ML across the industry at... at much larger scale than what you see today. 

Hannah Rosen: Are there any issues with artificial intelligence and machine learning that we just currently do not have the capacity to solve? 

Trevor Vieweg: I mean, I don't think we've properly solved the black box issue yet. I think that there are some interesting techniques coming out as to how to solve that, but that's... that's not a solved problem yet. You know, the other one is a big area of concern is, you know, eliminating bias from these models. And that's... that's a really tricky one because if you kind of follow that all the way back, a lot of the data that we have generated as humans includes a lot of bias in those models as well. And so that's a... that's a big talking point, particularly in life sciences, is... is how do you eliminate that? I'll say there, you know, I don't think you can eliminate that on day one. I think what you can do is, I think a better place to start is to say, make it no worse than the data that we have, and then slowly build that out of the model by improving, you know, how we're picking say, patients for a clinical trial or assessing patient outcomes by using AI and ML. So in the long run, I think you can use AI and ML to solve one of the main problems within there, but it's not something that you can solve right off the bat. And so those are... those are probably the two that I think of that are most fundamental to just AI and ML in general and the... and the life sciences space. 

Hannah Rosen: Yeah, I think that's a really great perspectives. I think that a lot of what you're saying right now is kind of emphasizing this point that, you know, AI and ML it's not a plug and play where like, you set it and forget it sort of thing. I suppose it's more of like this is something where you constantly have to be monitoring and working with it and shaping it and making sure that it is what you want. I almost think that, that could potentially ease some of the concerns people have surrounding AI and ML, is that it is an ever adapting process and it's not something that, you know, you just set it up and hope for the best. 

Trevor Vieweg: Yep, I... I think one thing that I've described with some friends, you know, that are less familiar with the space and... and have some concerns is, you know, the... the self driving car story, right. So 6-7 years ago some of the basic, I'll say, self driving capabilities came out from a number of different car manufacturers and, if you remember, a lot of articles came out basically saying that within two years humans will never have to drive again. Right. Sounded great, I said. We wanted that to happen, but I... I knew enough to know that that's a pretty tough problem to resolve, and what you see now I think is actually, you know, a good parallel to what you'll see in AI and ML, which is you see a huge amount of hype right now. You know, the... the killer robots are coming for us or something like that. But it's going to take a long time to get, you know, something that is truly artificial general intelligence. What you're going to get is a tool that's much better. So in the... in the self driving car world, what you get is autopilot and lane keep and variable cruise control and all these things that make your life as a driver way easier. I don't take road trips in any car nut my wife’s car now because she has all those features and mine doesn't. So, you know, it just makes your life much easier at that point. But I think that's what you're gonna see in AI and ML as well, is it's going to do all the unpleasant work for you. But, you know, we shouldn't think of it as something that's immediately solving every problem. And that, that actually cuts both ways, I think and the... the... the fears of it, but also into the promise of it. You know, we've... we've definitely had conversations with customers about wanting to use AI to solve, you know, every business and research problem they have. And, you know, we couldn't come back to say let's pick two, let's pick three, let's pick something that's very specific and a very painful point to start. And then we can grow out from there.  

Hannah Rosen: Are there any ways that we can accelerate the implementation of AI and ML solutions into the life sciences labs? 

Trevor Vieweg: I think education is great for researchers of just about what's possible and what's not. So I think, you know, some of the things that I've seen SLAS doing has been great. You know, about bringing some of that to the forefront with some of the talks there. I think that ChatGPT is actually kind of inadvertently being a great accelerator for getting people to think about how to be using these tools in the space. And, you know, and then the other... the other pieces are really, you know, what I was talking about before, which is finding ways to share data across groups and life sciences is going to continue to be a major challenge and barrier to adoption. So the faster that we can figure out how to share that data, yeah, I think the better off we'll be in using AI and ML. And then the... the last piece that comes to mind is really what we're working on in Limmi, which is bringing tools that are much more user friendly and make AI and ML accessible to folks that are, you know, not trained in machine learning and computer science. 

Hannah Rosen: You know, we spent most of this, or pretty much all this conversation focusing on, you know, the software side of things, for good reason. That's what your company focuses on, but I am curious, you know, as far as you're concerned, when we look to the lab of the future, you know, what differences do you see in the influence that advancements in laboratory hardware versus laboratory software are going to have on these lab spaces in the, what we all love to refer to, as the lab of the future. 

Trevor Vieweg: I'll use the self driving car analogy again, which is we were talking about the software side, but obviously there have been major advances that have enabled that, like much cheaper cameras, much cheaper radar, and LIDAR systems, hardware systems that really enable you to do what you're doing with software. I think the exact same thing holds true in the lab of the future on the hardware side, so you see major innovations in imaging systems in detection systems for, you know, various compounds or, you know, when it comes to genomic sequencing, ever finer detection mechanisms for protein expression and, you know, new NGS sequencers. All of that is complementary to the software side, and you're... you're going to get blocked. You know, you'll have this kind of back and forth where one side will get blocked by the other until the other side innovates. I think right now hardware has gotten much better than software in the space, but I want to see, you know, better... better hardware systems coming out because the software side can, you know, take advantage of that basically as soon as it's available. 

Hannah Rosen: Yeah, it does seem like, you know, of late, the focus when people talk about the lab of the future seems to have really kind of shifted away from conversations around hardware and it's more and more focused on these conversations around software. You know, do you think is this because the hardware has pretty much outpaced the software, or do you think that there's other factors at play? 

Trevor Vieweg: I think, you know, a lot of it comes back to just kind of the... the speed that you can discover, make new discoveries and, you know, the economics of it. So I think for a long time hardware was a bottleneck to that, right. I think the reason you see the shift is there is a huge amount of low hanging fruit in the software base when it comes to utilizing all the advancements that have come from the hardware side and processing that data, right. You know, you've got just a few examples, right? A lot of the next generation imaging systems, maybe not so much of like the research lab, but, you know, as you kind of move forward into patient diagnosis, right, you see all these iris images now that are wonderful for, you know, examining a patient's health and now, you know, what you see is this explosive software companies that are now taking advantage of that, right, to do AI detection on these images to do classification for cancers or for, you know, other diseases that humans may have. So I think that's why you see a lot of the shift towards the software conversation. I think people are just recognizing how much potential there is in the software space. If we can get there and get, you know, adoption there quickly, I think you'll see that shift again, you know, in a number of years where we've kind of exhausted a number of the possibilities of what we can do with the data, with the hardware technology we have today and need better and better hardware technology, I... I think the reality of that is the hardware technology will keep moving and you can just keep taking advantage of it on the software side. But I think that's why you see some of that conversation shift. 

Hannah Rosen: So when it comes to implementation of all of these solutions that you've talked about, you know, what differences do you see when it comes to implementation for larger, really well established companies versus smaller companies, or perhaps, you know, some newer startups? 

Trevor Vieweg: So this one actually looks a lot like the technology industry, I would say, which is that large companies tend to have a major data advantage. They have been around a lot longer, they probably run larger clinical trials. They just have a mountain of data to go exploit, you know, with an AI or ML. The challenge is it's stored in, you know, 30 different places in different formats. You know, maybe not up to pair with what you would want. You know, just in terms of kind of organization and labeling across the board, it's hard for those companies to, you know, get something in place to establish a program around that just because of the amount of data smaller companies, you know, are the flip side are learning to be much more AI native or computationally native, so they're getting these things set that stood up from the get go. We're working with a couple companies in this space. The challenge for them is then data access to generation, right? So almost by definition, they haven't generated the same amounts of data, and so the... the most successful ones that we see in the startup space are those that have a method to rapidly generate the data that they need, again, back to the hardware side, right? A lot of those hardware advances have made that a lot easier to do. And you know, with high throughput sequencing, customer sequencing, those sorts of things. The larger companies, I think the main challenge for them is just going to be getting that commitment and focus to, you know, making that change. And again, that's where I think we can help is... is by showing the power of some of these systems. So we're working on a number of demonstrations around use cases for this, hopefully to get larger companies to see the potential that they can access if they go make that commitment to, you know, accessing that huge amount of data that have. 

Hannah Rosen: Yeah, wonderful. We are about at the end of our time here today. Trevor, thank you so much for joining me and having this just really fruitful discussion, you know, surrounding AI and ML. I would encourage all the listeners out there to go check out Limmi and see if they have any solutions that could work for you. 

Hannah Rosen: And we really look forward to seeing Limmi at all of our future SLAS events. 

Trevor Vieweg: Likewise, Hannah. Again, appreciate the time and it was a lot of fun. So, look forward to talking with you again soon. 

 

Podcasts we love