Pharma Sessions
Pharma Sessions
Timothy Smith on AI, Data, and Drug Discovery at Takeda
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Pharma Sessions is hosted by Jonathan Kaskey
Follow along on LinkedIn: https://www.linkedin.com/in/jonkaskey/
Or join the Pharma Sessions Substack:
www.pharmasessions.substack.com
I think the key thing is focus on what your role is and look at it and say what things here are really repetitive that could easily be done by a machine. And go ahead and take advantage of the things like most companies now have some version of Copilot or Claude or Chat GPT and get ahead of it. You can automate some of those things and look at the part of your job that is really you can't think of a way to automate it and really demonstrate that value.
SPEAKER_01Hello, hello, and welcome to Pharma Sessions, a place for pharmaceutical leaders to come and learn from each other. I'm your host, Jonathan Kaskey. Technology and market trends are bringing change at an ever-accelerating rate, and no person, team, or company can afford to be left behind. Here, we dive into the strategies and tactics that our guests use to tackle these challenges and create new opportunities and how you can do the same in your own organization. This episode of Pharma Sessions is sponsored by Xunt, makers of the X1 reporting platform. On today's episode of Pharma Sessions, I'm thrilled to welcome Dr. Tim Smith, head of data science community and AI innovation at Takeda. Tim's career sits right at the intersection of science and technology, starting as a chemist and toxicologist to becoming an early adopter of machine learning in pharma long before the wave of AI hype that's currently in the middle of. He's helped build some of the first commercial AI applications in the industry, led large-scale collaborations with institutions like MIT, and now focuses on one of the hardest problems in pharma today, how to actually get people to adopt and trust AI at scale. What makes me particularly excited to talk to Tim is his perspective, it's really compelling about AI evolution from niche experimentation to enterprise priority. And now working on the human side of the equation, change management, trust, real world impact. So with that, we'll dive right in. Welcome to the show, Tim.
SPEAKER_00Thank you for having me. This is great.
SPEAKER_01Thank you. All right, before we get started, I always like to do a little icebreaker. And my current one, I used to ask people, what have you had to eat so far today? Which I feel like you always learn a lot about people. Uh, but now my current one is what is your go-to karaoke song?
SPEAKER_00My gosh. It's more important to know which one I'll never do again. Oh now I now we have to know. What is that? I'll meld with you. It's a song that I thought it would be easy to do. Yeah. And it went on forever. Yeah. And it was like, I just couldn't do it. So I think I like to do I Fought the Law. I love it. That's one of my favorite ones to do.
SPEAKER_01Clash is one of one of the greatest bands of all time. It's very, very good. Almost a little reggae inspired that one, too. So yeah, that's excellent. All right. So now we know a little bit about you, right? And tell me about your background. You started as a chemist and toxicologist. So how did you get from there to where you currently sit at AI and data science?
SPEAKER_00So the chemist piece early on, I left uh as a bachelor's, moved to San Francisco from Seattle, and was working at the University of California, San Francisco with a lab tech doing analytical chemistry. And I realized I didn't know nearly enough about the biology that I was supposedly doing the research on. So I thought I better go to grad school. And so toxicology seemed like the right mix of chemistry and biology. So I went and did a degree of PhD there in Cornell in neurotoxicology. And while in the postdoc following that work, I um was headhunted into a startup in Cambridge, Massachusetts, called Lion Bioscience. It was an early bioinformatics company. And that's how I transitioned. I sort of went from wet lab to in silico research. It was exciting. It was a really little company. It grew and then it grew too big and then disappeared. It's like an arc of about four and a half years, leaving me to uh pursue informatics in at a large pharma. I started at Novartis. I pretty quickly got, they realized I like to do sort of large-scale projects. So building like pipeline analytics, a database to track all of our projects across the company, you know, things like that. And in that process, I realized that there's something new on the horizon called machine learning. I was watching these Kagel competitions, and it was fascinating that the experts weren't doing as well as like predicting things like air traffic control, on time types of projects, things like that. And so I think this is something special that I gotta like take some time out of it.
SPEAKER_01Around what year is this that we're talking about?
SPEAKER_002008. Okay. Or nine, right around there. It was just kind of getting on my radar. I know, of course, it wasn't new, and they have like all the history and the AI winter, and it's been around for a long time, but this is when it kind of caught my eye. And we actually formed a little group at Novartists that we call them, I was called the Random Foresters, and we would meet and kind of talk about where we could apply this. And so that's kind of how it got started. And then in that process, we developed a few projects. Uh one of the first, I think the first commercial application of AIML in artists with a company called Who Knows. And it was an interesting idea of using combination of analytics, natural language processing, and behavioral analytics to see what people are working on, what they're reading, and then getting to know who they are so they could connect with other folks that are similar. It was an early project. It it didn't get additional funding. I think sometimes it's hard to show ROI off right off the bat. And so, but then in that process, I guess I got the fever and eventually I published a book.
SPEAKER_01Yeah, tell us about your book.
SPEAKER_00So the title is How to Profit and Protect Yourself from Artificial Intelligence. Yeah.
SPEAKER_01And I mean, I feel like that should be the number one bestseller right now these days, because those are the two things that everybody is both excited about and also kind of scared about, if if we're being honest. So what prompted you to take it to that level and to actually write a book about it?
SPEAKER_00I think it was that mixture of seeing how it's going to change pharma. And then I had gone down the street to a Friday afternoon seminar at MIT, and Eric Bernjolson was talking, and he was talking about how, you know, this new information revolution and how it was going to change jobs. And so I started reading more and more in that space and came across this really interesting idea of bias and and and also the idea of whose jobs are really at risk. And so that was kind of the profit and protect yourself initially. It was this is before LLMs and things like that came along. So we're really looking at automation. They already had like deep learning and things like that. So what's at risk? What can people do that machines can't do and when will that change? So it was kind of like exploring those concepts.
SPEAKER_01And what are some of the key themes that you were pulling out as far as what people should be doing in order to both, I guess we could take it one at a time. Let's start with avoiding the negative, right? In order to protect yourselves, what are some of the key themes that you're suggesting people should be considering?
SPEAKER_00In the book, we kind of explore initially a continuum of what are complicated things for machines to solve. And and so it could be, you know, at the either end of the spectrum is usually high creativity and then high physical dexterity. And so like things like hairdressers and construction workers, you know, working with a varying environment, they're far more protected than you would have for someone who does something that's in a controlled environment and is easily reproducible. So that was sort of the general theme. We've got to look to different jobs there. And I think the, as we're seeing now, even accounting and other things that used to seem like no way a machine could do it. That's based, and it's we're seeing uh more and more AI taking over that. And at the other end of the spectrum, the the highly creative, innovative thinking. Now I think we're gonna see machine assisted in that space, but I think that what is novel, the idea was at that point machine learning was really strong with lots of data. But the edge of knowledge, that sort of pointy edge, that's doesn't have a lot of data around it because that's the unknown. That's where people are are strong, you know, are stronger.
SPEAKER_01No, what if you're in a role like a lot of people that are listening to this are not in not in construction, are not in uh, you know, a creative artistic role. They might be in medical affairs, they might be in commercial operations, right? They might be doing sort of your standard office work at at Big Pharma. What is what is your advice to them short of changing careers?
SPEAKER_00I think the the key thing is focus on what your role is and and look at it and say what things here are really repetitive that I that could easily be done by a machine. And go ahead and take advantage of the things like most companies now have some version of Copilot or Claude or Chat GPT and get ahead of it. You can automate some of those things and look at the part of your job that is really you can't think of a way to automate it and really demonstrate that value. I it's it's it, you know, we're I just saw like Meta's laying off a ton of people in the news this morning. And there's some things I think we'll probably see some undulations where people will react and think, oh, AI is gonna do all these things for us, but then there'll be a retraction where certain things aren't working. I feel like, and I really relate with most people that your role in a company can kind of feel stable, it's not in the long run. When I was at Novartis, there was a guy and he just celebrated his 40th anniversary with the company, which was remarkable. That was when I first started there. And now you know that just seems impossible. Right.
SPEAKER_01It was actually it was in the news. The guy who I forget his name, who is taking over as CEO of Apple, his LinkedIn was very funny because it's like no picture and his two jobs or like an internship somewhere and then 26 years at Apple or something, yeah like that. That's right. But I think what you're saying, the way I think about this is operating at the top of your pay grade. And I'll give an example. Like, so if you are in sales at a pharma company, if you're a rep in the field, your job is to go out and have conversations with prescribers and ultimately drive prescriptions of your product within your territory. Your job isn't really to, but there's a lot of stuff you need to do in order to be good at that. That's not the point of your job, right? Like the point of your job is not to have a know your sales data and the marketing data and the responsiveness data to all your field-triggered emails or something like that, like the back of your hand, but it's really helpful to do that. So I think of AI as okay, well, what is the point of all of this is to do X. What's all the other stuff that I need to understand and I need to do in order to do X? And the more that you can offload some of that work, look at this data and come up, summarize what I should be taking from all of that, at least a first pass, and you can go back and you can pressure test it. You're putting time back in your day, and maybe that's how you go from seeing 2.5 doctors a day to 3.5 doctors a day. That's how you're increasing your value to the company. And I think there's a million examples like that, but to me, it all comes down to understanding really what is the point of your job, and then trying to you're almost delegating some of the stuff that that you need to do in order to be successful. It used to be to other teams, now maybe it's more to AI or it's a combination, but that's basically how I'm thinking about it. And I'm just curious your reaction to that.
SPEAKER_00That's a great example. I think the two things I might mention. One one thing that was brought up in or in the book I wrote was in the legal profession, certain things that are extremely automatable now. When you get, say, a million documents and you're doing corporate litigation, you used to have an army of people reading through that, looking for a sentence or a term or a concept that you are using to build your case. Now we can do that in seconds. And so those jobs are definitely going away. But it's the question, like, what am I actually looking for, which is the really high value thing now? And it kind of goes to what you're saying. Like, what are you really there for? And so they kind of went also then to case law. Certain things are boilerplate. And so they're really automatable. But being on the edge of deciding, you know, how to steer policy with changes in the law, these are human litigatable levels. And then also back then to pharma, one thing I really encourage people to do is we're all sitting in a lot of data. No matter what you do, your stack of emails is data. You're in a lab, you know, your instrument data, like it's all there. And now with these great tools like Cloud Code, you can really go in and start asking questions that you couldn't do before. You'd have to have a friend who is really knew how to code, who understood data science. You can really get a messy first run. And so being creative isn't always just about the person that knows how to write well or a person who knows how to compose a nice sonnet or something like that. Creative is what do I looking at? What is this could this tell me? Where your world will change if if you find something interesting. And people are more interested in following up too when when you have in data and a way to categorize it or summarize it or make predictions with it. That's all the great stuff I can do. And now it's closer to your finger than you can imagine.
SPEAKER_01If you've been enjoying the conversations here on Pharma Sessions, you should know they're made possible by the team at Xunt. Xunt helps life sciences companies turn complex data into clear, actionable insight. For years, Xunt has made complicated data sets simple to help commercial, medical, and operations teams map what's happening, predict what's next, and make stronger decisions faster. And now there's an added AI layer that makes everything work so much better. I was actually pretty jaded about some of the AI approaches I'd seen, but when XSunt showed me theirs, I actually left my job to come work for them. It's really awesome. So if you want to understand your market, your customers, or your performance with more clarity than ever, check out XSunt.com. That's XS-U-N-T.com. All right, let's jump back into the episode. So this is all very kind of relevant to the stuff that I see in the day-to-day, because just full disclosure, we have a data company and people are querying data with AI. But what's also interesting is it's you're sort of limited by your imagination now, right? Like it can tell you the things, but then if you're like, hey, add this to my dashboard, create a dashboard, it can actually do the work for you too. And those are things that used to be you had to go put in a log a ticket in Jira or something, and then your team would get back to you like a week later and you'd have it. And I see this all the time. Like our engineers are always saying, Bring us your ideas, right? And we'll spin up a mock-up and we'll have it to you this afternoon. Like it's not gonna be functional, it's not gonna be operable at scale, but you could at least play around with it and see if it's working the way you want. That was a bit of the avoiding the negative. Let's shift to the positive part and maybe then we can transition back into how do you then take that positive and actually implement it pharma, because I really want to get to that. But the title of your book is supposed to basically protect yourself in profit. So, so how if you're in these roles, where does the profit part come in?
SPEAKER_00So I think that's the by talking about like the protect yourself piece, you know, in terms of use these new tools, you know, don't don't wait, have someone else figure out how to use them, and then your your role becomes redundant. Yeah. The profit part of that is also look at what you're doing and the problems you're trying to solve and think about ways that AI can do that for you. And so it's kind of the same thing, but it it there's a number of different examples that we look at. And some of them are more not related to your day-to-day life in the sense of like, you know, looked, explored how deep learning, et cetera, was helping autonomous vehicles work and things like that. There's all around you, there's AI now, you know, in your phone and your car and everything else. But one example was kind of funny when I first came to Takeda, it was like the first day of COVID lockdown. And one of the things we were trying to do was understand how this is going to impact the company. And I'd only been there for like a week. And we had a ton of different types of data that we were looking at, but nobody was looking at the demographics, you know, in terms of the age. They were all looking about how often or how many people might be if infected. And so I always knew that the CIA has a ton of free data. You know, if you go to their website, it's a treasure trove.
SPEAKER_01CIA is in the Central Intelligence Agency?
SPEAKER_00Yes. Okay. Yeah. So you I I went there and they had all the demographic data for the world by country. And we also then had some internal data about the ages of people being infected. And it was really clear once you started mixing by country infection and by demographics, that like countries, Lesotho, which is like 50% under the age of 18, they almost had no infections. And then older countries had many more. And so it definitely was a at least a reportable disease at older levels. So that would be who we would want to look at first, you know. And so it was just this idea that and this is happening sort of in real in real time, right?
SPEAKER_01You're you're looking real time. Yeah. And you're basically you're coming up with a really key question, which is I guess even at the early stages, it sounds like we knew that it didn't impact everybody the same, that infection rates were different and outcomes were different. And then you were using a massive data, I mean world population data, to understand which countries were most at risk. And so, how long did it take you with your AI background, with your machine learning background? How long did it take you all to sort of parse that and come to conclusions?
SPEAKER_00The initial one was it just took like about a week. And working with because we had a really talented team of data scientists. And so I was, you know, kind of bringing things to the table again, like that first baked version. And then they were uh kind of a hybrid of data science, AI, and chemistry. I'm kind of partial master of some things. That's again like figuring out the problem. And then the creative part was well, where else can we get data? You know, it's not just like, well, here's our data set, what can we learn from it? Joining things is part of that creativity that it you don't have to be, like I said, like a Rembrandt or a Picasso. You have to just look around you and you know, notice there are things like I'm always curious and played around with for years the Sunshine Act data, which companies are paying which researchers and and kind of building trends and stuff like that from that data, knowing it's only a slice of the world, but it has a lot to tell you. And so you can then again join that with other types of data to come up with different insights.
SPEAKER_01It's interesting you use the word curious because as you were talking, I was thinking curiosity seems to be a really important attribute to have these days. And a big part of your role is around institutional change and enter enterprise change and adoption of this. So, what are you seeing around change management? What are the bottlenecks that you're trying to avoid? And how are you managing kind of bringing this to again a massive organization? Decade is a huge company at scale.
SPEAKER_00Yeah, I think I've seen is is some of the more kind of obvious things are they're adopted quickly because they're so easy to use. We saw from Chat GPT coming out in was it November of 2022, and by January it had millions of users. Right. It's because it was so easy. You went in, you typed some stuff, and it was pretty fantastic what you got back, even if it was wrong. You know, it's like it was wrong, but it was fun. But yeah, exactly. But I what I found with like enterprise applications, like I think that you have the strong side of Microsoft, which you know, most companies have that as their kind of background. Backbone for Excel and Word and document management, all those sort of things. That is pretty natural. Like people can use that and it works in that space. And it's that's Microsoft's strong suit. The other side of it is the research and development piece where you really are building unique applications for unique problems that are pharma specific. And these are the ones that are more challenging. And it's funny, what you're really trying to do is help the team that understands the problem really tell you what's the value. Not for their team. They intuitively know it. But then sometimes it takes even a few iterations over several weeks to get a narrative that that management understands, and then they'll fund going from your proof of concept to a production level product. And for me, that's one of the bigger challenges. And when it does work, um then it really goes because then everyone gets it. But it sometimes it can be so you get it viscerally, because that's why you've been doing this. But being able to translate it into something that that anyone understands in five minutes, that's kind of the next step.
SPEAKER_01Well, and then there's also even once they understand it, what I've seen, and I don't know if this is improving or not, but there are real sometimes operational challenges, even once everybody's agreed that this makes sense. And a lot of it is we keep referring to data as if it's this one cohesive thing, but a lot of times it is not even close to normalized. And it's like, even if you think about on the commercial side, right? Your prescription data is coming from IQVIA from Symfony, your activity data might be coming from Viva or Salesforce, then there's the pharmacy data, there's formulary data. Some of these might refer to patients as patients, some of them might say be PTS, period. There's all sorts of weird stuff that happens when you're combining data from different sources. Or like in other projects that I run, it's we're trying to, I used to do a lot of work in the documentation space. And the challenge with big companies is that they've acquired a million small companies and all of them have like slightly different formats and different templates and everything. And you're talking about like millions of documents. What's the chicken and the egg, right? When do you start looking at creating an underlying foundation that's gonna allow you to apply any type of AI or machine learning to it, or really probably AI at this point, versus you're creating the business case and then saying, all right, and now we need to go figure out how we're gonna actually give ourselves a resource that we can point this thing at?
SPEAKER_00No, I think that's a great point and a challenge. Over the years, have seen platforms come and go, you know, especially they're sometimes easier to build and fund because they have that almost five-minute elevator pitch built in. You know, it's sort of like, let's build a data lake and it'll be able to do X, Y, and Z. And then several years later, and millions of dollars, it's sort of like, oh, we didn't really do the right metadata. And so it's now it's got everything in there, but we can't really find it. And so that's, you know, that is kind of that is 100% accurate, right?
SPEAKER_01That is uh very true. I'm sure that's a true story.
SPEAKER_00It's not a criticism per se, but it's sort of like when you build up from a small problem, you kind of grow. But when you have like a big solution and then hope it finds problems, it can be a challenge. But I think one of the things that I think is a really good idea is really understanding where your data is and what it is. And and that's kind of a challenge. It's not impossible. Financial data generally is financial data, but it probably has a few hooks in it that connect it to your your pipeline, for example. And your pipeline then should have what targets you're working on. And so there are threads through all of this that can make them connectable and and might not always be the fastest or best thing is to build a big platform and then you know, after several years, turn on the questions. I'm okay with a little bit messy, yeah, where you know some of the things are working really smooth, other ones are gonna take some work. You know, one of these problems we had was trying to build a system where you can model data and you can borrow data from other labs. It's similar. And that was a a cool idea, but I think one of the problems was even slight differences in the data made it a little trickier to make the comparisons that you thought you might make. Yep. But the best part was that you could actually see the other labs' data. So say a bunch of people were doing flow cytometry, you could see these guys are doing that, this other lab's doing it, but their analytics might be very different. And that's where the conversation was actually helpful. The tool didn't do it, uh it was kind of a not a success, but that conversation was a success. Yeah, you know, so kind of pointed to, well, should we all use the same, you know, vendor or should we all build something internally that does exactly what we want? You know, these are, I think, important ways for people to connect. And that kind of goes back to how you profit. I think people are still the key to all how all this works right now. Because some someone has to make sense of it and then propagate it is crucial. Someone might say, well, there'll be some super smart agent. One day it'll just make all these decisions on its own. And I don't know, maybe.
SPEAKER_01Right. Right. Let's close with just a little bit of forward looking. But if if you were going, let's say you get pulled into a room and you've got, you know, all the pharma CEOs in the world, they're all listening to you, they're all hanging on your every word. Um, what's the one mindset shift that you would suggest they take to really position themselves and their companies up for success going forward in this world of AI?
SPEAKER_00I think that particularly in pharma, and again, I have the RD hat. That's kind of where I come from, is are taking advantage of generative AI for really everything from target selection up through the clinic, the more integrated that that knowledge set is, the faster and better you're gonna work. And the the caveat to that is a friend of mine who's you know of a company that does a lot of automated compound discovery, even he said, you really, really depend on the medicinal chemist still. He said, I thought I could automate them out of the process. You know, and so really understanding where you have and need expertise and cultivate that expertise, pay for it, because you can generate a ton more molecules now in a shorter amount of time. But having some people that can comment on, you know, are you on the right track? Yep, is hyper valuable. And that's where I'm still very um people-centric. And I think that's where also you want to develop your people that are basile, they want to work with the assistants, but they have an opinion and they have over time uh an expertise that helps them be critical.
SPEAKER_01Yeah, I think that's a great point, right? Because at the end of the day, pharma is its own unique beast. Like if you were trying to, let's say you were cre at a creative agency and you're making, you're making an advertisement and you come up, you use AI and say, okay, well, this gets us, this gets us 80% of the way there, and it's faster and it's cheaper. That might be okay. But if you're creating a drug that's about to have to go through a 10-year clinical trial at the cost of hundreds of millions of dollars, it's probably not good enough, right? So so the idea is is like get there faster, get there faster, but you still do ultimately need something that is really held to a different quality standard than a lot of other industries for good reason, right? There's a we we all depend on these, or we will at some point in our lives. So, okay, Tim, thank you so much. This was I had a lot of fun. Um, I will publish the link to your book when we post the podcast episode. So anybody who's made it at this point, thank you and uh check out the show notes and read Tim's book. I'm going to. Thank you so much, Jonathan. This is a pleasure. All right, great, excellent. This episode of Pharma Sessions is sponsored by Xunt, makers of the X1 reporting platform. And that's a wrap on today's episode of Pharma Sessions with me, Jonathan Kaske. If you enjoyed today's conversation, don't forget to hit follow or subscribe and share it with someone else in the pharma world who might need to hear it. For more on pharma trends, career growth, and business strategies, connect with me, Jonathan Kaske, on LinkedIn. Until next time, thanks for listening.