Ep. 323 - Agentic AI: From New Targets to the Clinic

WEBVTT

00:00:00.000 --> 00:00:02.201
[AI-generated transcript.]

00:00:02.201 --> 00:00:05.671
<v Eric Pierce>BioCentury This Week is brought to you by IQVIA Biotech.

00:00:06.506 --> 00:00:15.048
<v Eric Pierce>For biotech companies striving to bring innovative therapies to market and maximize patient impact, IQVIA Biotech is the trusted CRO of choice.

00:00:15.281 --> 00:00:25.658
<v Eric Pierce>Backed by 25 years of unparalleled experience and deep therapeutic expertise, our full-service clinical development solutions are purpose-built to accelerate success.

00:00:25.958 --> 00:00:35.402
<v Eric Pierce>IQVIA Biotech helps early-stage biotechs de-risk by developing strategic clinical development plans, guiding drug candidates along the most promising pathways.

00:00:35.734 --> 00:00:44.243
<v Eric Pierce>Leverage data-driven models and dynamic tools to craft a compelling value story and maintain momentum through every phase of drug development.

00:00:49.182 --> 00:00:56.155
<v Jeff Cranmer>AI is the buzzword everywhere these days and biotech is no stranger to that.

00:00:56.622 --> 00:01:01.895
<v Jeff Cranmer>Welcome to a special edition of the BioCentury This Week podcast.

00:01:02.529 --> 00:01:04.496
<v Jeff Cranmer>Today I'm very pleased.

00:01:04.697 --> 00:01:13.673
<v Jeff Cranmer>we have a special guest from our sponsor, Greg Lever, he's the Director of AI Solutions Delivery at IQVIA.

00:01:13.939 --> 00:01:22.781
<v Jeff Cranmer>And joining me as well are my colleagues and podcast, regulars, Selina Koch and Lauren Martz.

00:01:22.781 --> 00:01:28.787
<v Jeff Cranmer>And we're gonna cut right to the chase, Selina, when you think of AI in biotech, uh, what does it say to you?

00:01:29.822 --> 00:01:39.165
<v Selina Koch>right now if I try to take a step back and like take in all these changes that are sweeping through the industry so rapidly, it kind of makes my head spin a little bit.

00:01:39.766 --> 00:01:51.144
<v Selina Koch>conversations about,  how to use AI and agentic AI and drug development, are now, you know, being happening in, so it seems like in every single company, at every investment firm, and it's even.

00:01:51.844 --> 00:01:54.847
<v Selina Koch>Maybe even hard to go to a meeting where the topic doesn't come up right?

00:01:54.847 --> 00:02:13.032
<v Selina Koch>So things are changing at kind of dizzying speed, but because of that, we're in this like moment where there's this wide variability across the industry in terms of how much or little people understand about the technology and how proactive they are in engaging with it, I think.

00:02:13.299 --> 00:02:16.902
<v Selina Koch>And so I think it's a really nice time to have this conversation to kind of take stock.

00:02:17.169 --> 00:02:24.377
<v Selina Koch>Of where agentic AI is, making some practical inroads and, where it has, other applications in the future.

00:02:24.743 --> 00:02:30.516
<v Selina Koch>if I could say one more thing, I don't think that's always obvious, like where AI is gonna make contribution.

00:02:30.817 --> 00:02:37.356
<v Selina Koch>like I don't think many people would've guessed that writing code would be among the first breakthrough applications of AI that.

00:02:37.657 --> 00:02:40.560
<v Selina Koch>First jobs under threat would be programming jobs, for example.

00:02:40.560 --> 00:02:46.265
<v Selina Koch>But you know, in retrospect it makes sense when you think about the data sets that are available in that space.

00:02:46.532 --> 00:02:54.340
<v Selina Koch>There are, well, I don't know, or Greg might know this, um, tens of millions of programmers who regularly check code into GitHub.

00:02:54.540 --> 00:03:02.414
<v Selina Koch>So there's this really big repository that not only has a lot of code, but has like a code progression, leading to code working or not working.

00:03:02.414 --> 00:03:03.015
<v Selina Koch>So you have.

00:03:03.483 --> 00:03:05.919
<v Selina Koch>Also this reference to ground truth.

00:03:06.286 --> 00:03:09.722
<v Selina Koch>but when I think about biology, it's really not that straightforward, right?

00:03:10.156 --> 00:03:13.293
<v Selina Koch>We have this like fragmented evidence ecosystem.

00:03:13.693 --> 00:03:18.131
<v Selina Koch>We have a few large scale data sets would have that kind of reference to ground truth.

00:03:18.364 --> 00:03:22.534
<v Selina Koch>that makes me question in the near term where there's gonna be practical application.

00:03:23.002 --> 00:03:25.405
<v Selina Koch>Like if you think of protein design, for example.

00:03:25.705 --> 00:03:32.211
<v Selina Koch>I can imagine that optimization of antibodies and proteins along certain parameters will, will be happening.

00:03:32.211 --> 00:03:33.346
<v Selina Koch>It is happening, right?

00:03:33.346 --> 00:03:36.782
<v Selina Koch>But wholesale de novo design could take a while longer.

00:03:36.816 --> 00:03:37.282
<v Selina Koch>I dunno.

00:03:37.750 --> 00:03:39.451
<v Selina Koch>If you all disagree, feel free to chime in.

00:03:39.451 --> 00:03:39.618
<v Selina Koch>But

00:03:40.052 --> 00:03:43.389
<v Jeff Cranmer>Lauren, do you have any thoughts that you wanna throw out there before we, uh, go?

00:03:43.389 --> 00:03:47.860
<v Jeff Cranmer>I see Greg nodding along and we'll, we're gonna bring him in in just a minute, but, uh, Lauren.

00:03:48.228 --> 00:03:57.402
<v Lauren Martz>Yeah, I think we can pass it to Greg momentarily, just to follow up on some of the ways that this technology might be finding its way into our industry and already is.

00:03:57.669 --> 00:04:00.740
<v Lauren Martz>there's protein design, there's target discovery.

00:04:01.707 --> 00:04:07.447
<v Lauren Martz>And that goes all the way through to ways that, AI can support clinical trial design.

00:04:07.713 --> 00:04:10.250
<v Lauren Martz>you know, we've been hearing about this for years.

00:04:10.483 --> 00:04:12.551
<v Lauren Martz>I think a lot of what we've heard initially is that.

00:04:13.252 --> 00:04:19.992
<v Lauren Martz>This can be used to help with site selection, you know, with, finding the locations and all the different dimensions that go into that.

00:04:20.059 --> 00:04:22.028
<v Lauren Martz>This is a place where data can have a big impact.

00:04:22.028 --> 00:04:28.134
<v Lauren Martz>You know, where the patients are, where the right investigators are, um, you know, how many trials are being run in that, in that site.

00:04:28.166 --> 00:04:40.747
<v Lauren Martz>This is something that we, we've already seen this in action, from Greg, I'd love to hear more about how AI can support clinical trial design and everything in this spectrum of drug discovery, through drug development.

00:04:41.346 --> 00:04:43.048
<v Jeff Cranmer>Alright, well let's bring in Greg.

00:04:43.048 --> 00:04:48.053
<v Jeff Cranmer>It's Greg Lever, director of AI Solutions Delivery at IQVIA.

00:04:48.588 --> 00:04:57.663
<v Jeff Cranmer>And Greg, I know,  we've chatted a bit before and I think you thought it was important to kind of set some definitions to get us oriented.

00:04:57.829 --> 00:04:59.197
<v Jeff Cranmer>why don't I hand it over to you?

00:04:59.798 --> 00:05:00.132
<v Greg Lever>Thank you.

00:05:00.132 --> 00:05:00.899
<v Greg Lever>Yeah, thanks for having me.

00:05:00.899 --> 00:05:01.733
<v Greg Lever>It's great to be here.

00:05:01.767 --> 00:05:18.350
<v Greg Lever>Yeah, and I think to Selina's point, there's a lot going on in the industry and I think it can be quite difficult to get a, a proper grip on a lot of the terminology that's out there and the different approaches and, and also why they matter and, and, and what it is that we can, we can actually, you know, use here.

00:05:18.350 --> 00:05:22.355
<v Greg Lever>So I think to act as a, a starting point, it can be, helpful to think about.

00:05:22.622 --> 00:05:39.439
<v Greg Lever>How all the different terms interconnect is specifically aspects like ai, machine learning, generative ai, ai, and if we start with ai, as a broad term, artificial intelligence is really, you know, used to describe all of these sorts of initiatives.

00:05:39.439 --> 00:05:43.408
<v Greg Lever>And it was, started as an academic discipline in the mid 1950s.

00:05:43.675 --> 00:05:47.947
<v Greg Lever>and there are now vast different types of approaches and algorithms within this space and.

00:05:48.348 --> 00:05:51.184
<v Greg Lever>One of these approaches is, machine learning.

00:05:51.184 --> 00:06:01.593
<v Greg Lever>So this is where, you know, typically you have algorithms that can be used to train a model so that it can learn from data, make predictions based on unseen, but related data.

00:06:01.593 --> 00:06:08.000
<v Greg Lever>So, as a more concrete example, say I have data on adverse events that occurred in various clinical studies.

00:06:08.000 --> 00:06:14.607
<v Greg Lever>You can build a machine learning model to predict the probability of adverse events of studies that the model hasn't previously seen.

00:06:15.040 --> 00:06:19.278
<v Greg Lever>But there are limits here to this in terms of only being able to answer this specific question.

00:06:19.312 --> 00:06:21.180
<v Greg Lever>It needs very specific data.

00:06:21.413 --> 00:06:25.752
<v Greg Lever>but what's been really impactful more recently is generative ai.

00:06:25.752 --> 00:06:26.853
<v Greg Lever>So this is where.

00:06:27.420 --> 00:06:38.130
<v Greg Lever>The model can generate new outputs, in the form of video, audio, or language like we see with large language models, you know, also referred to as LLMs.

00:06:38.463 --> 00:06:45.137
<v Greg Lever>And it's these sorts of approaches that are made possible by the work that was done on transformer models.

00:06:45.637 --> 00:06:48.641
<v Greg Lever>and so, you know, there was a landmark paper from 2017.

00:06:48.908 --> 00:07:03.555
<v Greg Lever>A year later we saw the first release of the generalized pre-trained transformer or GPT model, and then later iterations of this is what were launched in 2022 in the form of ChatGPT that was, you know, took the public by storm on the internet.

00:07:03.555 --> 00:07:08.494
<v Greg Lever>And so LLMs like those, powering these chat bots like ChatGPT.

00:07:08.927 --> 00:07:17.836
<v Greg Lever>What they're doing is they model language as a sequence of word fragments or, or tokens, and generated token by token, it's this new text.

00:07:18.437 --> 00:07:19.204
<v Greg Lever>That's coming out.

00:07:19.204 --> 00:07:22.809
<v Greg Lever>It's based on all of the preceding text that has gone into the model.

00:07:22.809 --> 00:07:35.321
<v Greg Lever>And so with enough training, these statistical dependencies among these tokens, this interconnectedness, it proves sufficient to actually produce what we see as conversational text.

00:07:35.321 --> 00:07:39.625
<v Greg Lever>And so, you know, often it's basically indistinguishable from that of a human counterpart.

00:07:39.992 --> 00:08:02.949
<v Greg Lever>the link here for something like clinical development prediction is that in the same way that language model learns from, you know, the grammar, the contextual logic of language from vast bodies of, basically internet scale text data sets, clinical development models based on these same types of approaches can begin to infer patterns of progression.

00:08:03.149 --> 00:08:05.350
<v Greg Lever>Um, you know, when they're trained on data from things like.

00:08:05.651 --> 00:08:09.988
<v Greg Lever>Preclinical readouts, clinical studies, approval documents, just to name a few.

00:08:10.490 --> 00:08:19.665
<v Greg Lever>both types of approaches can recognize these past events, but exploit these dependencies, this interconnectedness to predict future sequences.

00:08:19.665 --> 00:08:22.634
<v Greg Lever>So that's whether, what's the next word in the sentence?

00:08:22.869 --> 00:08:29.141
<v Greg Lever>based on a question that's been asked in a prompt in ChatGPT or say the next milestone, in a clinical program.

00:08:29.742 --> 00:08:37.115
<v Greg Lever>And so when we think about these language models, that there's also another aspect within, within AI that we're hearing a lot about at the moment.

00:08:37.450 --> 00:08:39.184
<v Greg Lever>and this is agentic ai.

00:08:39.184 --> 00:08:48.326
<v Greg Lever>And you know, really, really, if we think about a quick definition here, agentic AI is talking about AI systems that can act with agency.

00:08:48.326 --> 00:08:51.163
<v Greg Lever>And what that means is that they have this autonomous.

00:08:51.596 --> 00:08:58.303
<v Greg Lever>ability to analyze data, but also make decisions, execute tasks, if that's what you need them to do.

00:08:58.303 --> 00:09:03.241
<v Greg Lever>So, unlike traditional machine learning or generative AI approaches.

00:09:03.609 --> 00:09:10.216
<v Greg Lever>These agents, they're designed to reason, but also to plan and act independently on these plans.

00:09:10.549 --> 00:09:17.023
<v Greg Lever>And basically this gives this whole new level of automation, and adaptability and especially within clinical research.

00:09:17.056 --> 00:09:24.263
<v Greg Lever>So unlike traditional language-based, applications, I love that I'm talking about traditional language models as if this has been going on for.

00:09:24.864 --> 00:09:25.431
<v Greg Lever>Decades.

00:09:25.664 --> 00:09:28.100
<v Greg Lever>Um, and you can see how fast this is moving.

00:09:28.100 --> 00:09:36.509
<v Greg Lever>So now you can dynamically choose tooling to incorporate maybe reasoning, adapt their analysis based on the situation at hand.

00:09:36.841 --> 00:09:42.914
<v Greg Lever>And really what differentiates, agents from, from more conventional AI approaches is, is.

00:09:43.316 --> 00:09:45.083
<v Greg Lever>Really, it's a combination of four things.

00:09:45.116 --> 00:09:48.854
<v Greg Lever>They've got an an underlying model that they're utilizing.

00:09:48.854 --> 00:09:54.894
<v Greg Lever>So they might have a language model, they might have a reasoning model, essentially something that serves as like that brain.

00:09:55.528 --> 00:09:57.263
<v Greg Lever>They also have tools.

00:09:57.330 --> 00:10:09.708
<v Greg Lever>AI agents have things like maybe, um, databases or, or APIs, other tooling, maybe even conventional machine learning models that they can use,  to inform, the insights they're bringing back.

00:10:10.076 --> 00:10:15.147
<v Greg Lever>But they also have, um, they have memory, they have additional information that's been brought in.

00:10:15.480 --> 00:10:22.621
<v Greg Lever>And the key thing is, logic that helps the agent figure out what do they do next based on that current state and these decisions.

00:10:22.621 --> 00:10:35.167
<v Greg Lever>So really to, to finish that up, the analogy to think of with agentic approaches is, A language model can, if you ask it to recommend the best places to visit in Tokyo, it will do that for you.

00:10:35.167 --> 00:10:36.302
<v Greg Lever>And it do, do it pretty well.

00:10:36.568 --> 00:10:38.604
<v Greg Lever>an agent can do that as well.

00:10:38.604 --> 00:10:43.509
<v Greg Lever>But it can also book your flights, your accommodation, make dinner reservations, you know, book you into a show.

00:10:44.043 --> 00:10:47.013
<v Greg Lever>these are the sort of autonomous aspects that we're then seeing.

00:10:47.113 --> 00:11:00.393
<v Greg Lever>And you know, as you can imagine, within life sciences, healthcare, clinical trials, there are, there are huge amounts of guardrails and other approaches that we need to make sure are in place before we just, you know, let these agents go off autonomously and do what they want.

00:11:01.159 --> 00:11:02.227
<v Selina Koch>that was super interesting.

00:11:02.461 --> 00:11:16.307
<v Selina Koch>so when it comes to the scope of agents, I guess, do you think it's best when an agent is, designed to have a very specific function that it fills and then you string together different agents?

00:11:16.708 --> 00:11:22.682
<v Selina Koch>With some sort of orchestrator or how are people thinking about building more complex end-to-end systems?

00:11:23.916 --> 00:11:25.583
<v Greg Lever>Yeah, no, that's exactly right.

00:11:25.618 --> 00:11:33.158
<v Greg Lever>The the real key advantage that we see with agentic frameworks is that you go beyond this.

00:11:33.658 --> 00:11:41.966
<v Greg Lever>Initial situation you might have seen, say five, 10 years ago, where to my example, before I've built a machine learning model that can predict adverse events.

00:11:41.966 --> 00:11:44.202
<v Greg Lever>I have this very specific question and then that's it.

00:11:44.403 --> 00:11:45.370
<v Greg Lever>I can't go beyond that.

00:11:45.671 --> 00:11:58.884
<v Greg Lever>Whereas if I,  build my very specialized, framework of agents, some which may have access to data, about adverse events, some may have data on, um, preclinical readouts approval documents.

00:11:59.150 --> 00:12:03.288
<v Greg Lever>But they also have those language models so that yes, they can bring those insights in.

00:12:03.589 --> 00:12:11.029
<v Greg Lever>And just like you said, Selina, there's, there'll be some orchestrator agent, which not only, brings in the insights from the different agents, but also.

00:12:11.330 --> 00:12:29.981
<v Greg Lever>Fully understands the intent of the question from the user so that now I'm at a state where, yes, I can predict if there's gonna be an adverse event in an upcoming clinical study, but actually I can go beyond that and answer additional questions that we didn't have to bake in at the start of designing this whole framework.

00:12:29.981 --> 00:12:36.922
<v Greg Lever>And that's, that's kind of some of the magic and the power that comes out of these things and, and actually how we utilize this in the best way.

00:12:37.355 --> 00:12:39.692
<v Greg Lever>That's still not a completely solved problem.

00:12:39.692 --> 00:12:42.695
<v Greg Lever>That's the exciting piece about, agentic frameworks at the moment.

00:12:43.629 --> 00:12:45.364
<v Jeff Cranmer>Let's bring that to target discovery.

00:12:45.364 --> 00:12:47.232
<v Jeff Cranmer>What can and can it do in target discovery?

00:12:48.167 --> 00:12:49.902
<v Greg Lever>No, I think, I think that's really interesting.

00:12:49.902 --> 00:12:59.378
<v Greg Lever>It's, where these source of approaches, you know, they're currently being applied,  in, target discovery in that earlier, piece of the discovery phase.

00:12:59.644 --> 00:13:01.013
<v Greg Lever>especially when you're looking for.

00:13:01.413 --> 00:13:06.217
<v Greg Lever>Opportunities for selecting or identifying, um, novel targets.

00:13:06.217 --> 00:13:11.524
<v Greg Lever>You know, we know that those existing pain points are, how do we, how do we identify those novel targets?

00:13:11.524 --> 00:13:13.993
<v Greg Lever>We've got all this fragmented and disconnected data.

00:13:14.025 --> 00:13:18.663
<v Greg Lever>We, we might have limited resources for actually doing a really deep landscape analysis.

00:13:18.964 --> 00:13:25.937
<v Greg Lever>And actually there's just a huge risk of missing opportunities, and especially maybe in rare or emerging indications.

00:13:26.304 --> 00:13:31.844
<v Greg Lever>Um, and so there's a couple of approaches that IQVIA uses to support these sorts of challenges.

00:13:31.844 --> 00:13:39.217
<v Greg Lever>And so really you can think about it, is taking a language model that's been trained on an internet scale dataset.

00:13:39.451 --> 00:13:44.089
<v Greg Lever>you can think about this as a library and you can think of a standard language model like.

00:13:44.590 --> 00:13:49.195
<v Greg Lever>Someone who's read this entire library, they can recall what they've memorized during the time they read it.

00:13:49.195 --> 00:13:55.134
<v Greg Lever>If they're asked a question, but it's based on what they remember, that might be a little bit incomplete or it might be outdated.

00:13:55.167 --> 00:14:00.739
<v Greg Lever>And so the first step in going beyond this is, is known as what's called, um, retrieval, augmented Generation or RAG.

00:14:01.307 --> 00:14:09.748
<v Greg Lever>Basically, you can imagine giving person like a librarian to point out here are the really useful parts of the library that are relevant for the question that you have.

00:14:10.081 --> 00:14:16.522
<v Greg Lever>and, and maybe there might be updated materials that the librarian can bring in that the person didn't have originally available to them.

00:14:16.922 --> 00:14:21.759
<v Greg Lever>And then the second step is to go beyond this, um, using an approach called GraphRAG.

00:14:22.027 --> 00:14:31.903
<v Greg Lever>Basically what this does is it creates a knowledge graph so that in practice, um, you know, now imagine the librarian is even more sophisticated.

00:14:31.903 --> 00:14:37.842
<v Greg Lever>They can, help the person understand these are the important sections of the library, but these are the most relevant books.

00:14:37.942 --> 00:14:42.581
<v Greg Lever>And also these are the relationships between concepts across these books.

00:14:42.581 --> 00:14:50.623
<v Greg Lever>Like for example, how, you know, ideas in one chapter might influence another, how different authors discuss the same point from different angles.

00:14:50.855 --> 00:14:58.464
<v Greg Lever>it allows your language models generate not only, um, informative answers, but ones that are, you know, contextually rich and connected.

00:14:59.097 --> 00:15:06.804
<v Greg Lever>And so the agentic piece, although also comes in when we use these specialized agents to extract data from different places.

00:15:06.804 --> 00:15:12.577
<v Greg Lever>It might be scientific literature, trial registries, approval documents to name a, name, a few.

00:15:13.011 --> 00:15:16.916
<v Greg Lever>We identify these underlying entities within and across data sets.

00:15:17.249 --> 00:15:28.793
<v Greg Lever>And essentially what it, what it allows you to do is have this optimized language model where now you can start answering questions like, what are the future trends expected in a particular disease area?

00:15:28.793 --> 00:15:35.234
<v Greg Lever>What are the potential growth areas to consider, within my TA of interest, you know, are there some indications that show.

00:15:35.500 --> 00:15:38.504
<v Greg Lever>more promise of their pathologies or their mechanisms of action.

00:15:38.504 --> 00:15:48.246
<v Greg Lever>So really,  it's the sort of thing that may enable a biotech to move beyond just, you know, incremental innovation, and actually pursue some, some truly, novel approaches.

00:15:48.681 --> 00:15:59.557
<v Selina Koch>So on the target discovery front, when it comes to ingesting all of those different kinds of data, making sense of it, like what would be some shorter term benchmarks of success that you would be looking for?

00:16:00.525 --> 00:16:02.360
<v Greg Lever>And I think this has to really.

00:16:02.727 --> 00:16:08.833
<v Greg Lever>Align with your metrics for success in terms of your overall R&D pipeline.

00:16:08.833 --> 00:16:18.409
<v Greg Lever>And so, if that's, if there's, if there's maybe kind of more, more assets that you've got in mind and it's, it's always gonna be these longer timescale, considerations of.

00:16:19.077 --> 00:16:19.510
<v Greg Lever>Right.

00:16:19.778 --> 00:16:26.784
<v Greg Lever>This is what I'm looking at to say de-risk my R&D pipeline for the next say, 10 years.

00:16:26.784 --> 00:16:36.828
<v Greg Lever>If I'm looking at a 2035 roadmap, these are the mechanisms of action or these are the interesting, kind of pathologies or indications that I need to be identifying now.

00:16:37.261 --> 00:16:42.900
<v Greg Lever>Potentially there's also opportunity to course correct, um, existing R&D pipelines.

00:16:43.034 --> 00:16:48.807
<v Greg Lever>you might need to optimize in some way, but I think that's where you might think about a different type of approach.

00:16:48.974 --> 00:16:53.245
<v Greg Lever>maybe if you're thinking about probability of technical and regulatory success.

00:16:54.013 --> 00:16:55.447
<v Jeff Cranmer>That's fascinating stuff, Greg.

00:16:55.447 --> 00:16:57.682
<v Jeff Cranmer>I'm curious what, what's next?

00:16:58.182 --> 00:17:13.565
<v Greg Lever>Exactly, so, so while this, this knowledge graph creation work is the sort of thing that's happening right now as agentic frameworks mature and we begin to expand on their capabilities, you can imagine agents completing tasks like, okay.

00:17:14.333 --> 00:17:16.567
<v Greg Lever>Go through this knowledge graph that's available to you.

00:17:16.835 --> 00:17:21.339
<v Greg Lever>Identify indications where an asset may not have been successful.

00:17:21.507 --> 00:17:22.875
<v Greg Lever>Based on data available.

00:17:23.208 --> 00:17:32.917
<v Greg Lever>You might be able to utilize existing subpopulation analysis models to predict where patient subtypes may in fact respond well to a previously failed asset.

00:17:33.184 --> 00:17:37.588
<v Greg Lever>And you know, this, this supports things like defining eligibility criteria for future studies.

00:17:38.190 --> 00:17:44.863
<v Greg Lever>In a similar way to looking at, you know, approved therapies and identifying potential areas of expansion.

00:17:45.196 --> 00:17:56.541
<v Greg Lever>And this is something that drug repurposing as a concept has been working on for, for a while, but it's really going to be accelerated by agents being able to work autonomously over this data.

00:17:56.942 --> 00:18:01.346
<v Greg Lever>But also being creative about how to bring back these insights.

00:18:01.346 --> 00:18:12.356
<v Greg Lever>And I think that's something that often gets underestimated is the language models capabilities for, for creativity, but also being mindful that that doesn't turn into hallucinations, which

00:18:12.391 --> 00:18:16.528
<v Selina Koch>That was what I was just gonna ask you about how, how do you be mindful that it doesn't turn into hallucinations?

00:18:16.528 --> 00:18:18.463
<v Selina Koch>I mean, you wanna use positive controls.

00:18:18.463 --> 00:18:22.901
<v Selina Koch>I assume, like any experiment does it tell me the things I expect to see, but are there other tips?

00:18:24.169 --> 00:18:41.086
<v Greg Lever>So I think really one of, one of the things that's, that's crucial to thinking about is  in the same way that you have to utilize various metrics and key performance indicators, when you think about any kind of, predictive model, whether you've just put something together in Excel, or you've got a simple machine learning model.

00:18:41.519 --> 00:18:50.162
<v Greg Lever>And agents have exactly this as well, and so large language models will have things like temperature and other parameters and aspects that you can look at to test.

00:18:50.528 --> 00:18:54.432
<v Greg Lever>Whether it's hallucinating or whether it is genuinely being creative.

00:18:54.799 --> 00:19:04.509
<v Greg Lever>And I think one example to think about is you can think of this future where, you might have a, a biotech with an interesting asset or a particular mechanism of action.

00:19:05.042 --> 00:19:17.823
<v Greg Lever>There might actually be a portfolio of, say, rare diseases with a shared mechanism, which can be impactful over multiple, relatively smaller patient populations and thereby making the asset more attractive to investors.

00:19:18.022 --> 00:19:19.223
<v Greg Lever>And it's the sort of thing that.

00:19:19.691 --> 00:19:22.426
<v Greg Lever>Right now is a very manual effort.

00:19:22.426 --> 00:19:39.076
<v Greg Lever>We'll take a lot of research and a lot of, a lot of work to generate this sort of hypothesis, but it's the sort of aspect that if you have an agent that has this key task, but it's been set off to do, um, it can make it, it can make it much, much more easier to bring back in those sorts of insights.

00:19:39.478 --> 00:19:44.982
<v Selina Koch>That's a very optimistic example given the, uh, barriers to the business case in rare diseases.

00:19:45.017 --> 00:19:45.683
<v Selina Koch>so I like that one.

00:19:46.285 --> 00:19:46.551
<v Selina Koch>Okay.

00:19:46.551 --> 00:19:53.424
<v Selina Koch>Well, we've talked some about, the applications, you know, near and short term in clinical development and those, for target discovery.

00:19:53.424 --> 00:19:55.193
<v Selina Koch>Well sandwiched in between those two things.

00:19:55.193 --> 00:19:59.765
<v Selina Koch>There's designing a molecule, predicting whether or not it's going to be successful.

00:20:00.365 --> 00:20:01.666
<v Selina Koch>let's dig into that a little bit.

00:20:02.733 --> 00:20:03.335
<v Greg Lever>Absolutely.

00:20:03.335 --> 00:20:11.676
<v Greg Lever>So I think one of the things to think about, in terms of some of the forward looking future thinking aspects as well, is that.

00:20:12.243 --> 00:20:32.364
<v Greg Lever>Once a, a target has been identified, the sorts of things that maybe early stage biotechs can think of is that AI agents are going to increasingly be used to assess things like technical, regulatory, operational success, and really how they can reduce their risk and increase confidence in their, in their development strategy.

00:20:32.364 --> 00:20:47.945
<v Greg Lever>And so the way, the way these sorts of things will work is there'll be these specialized agents that will be able to understand aspects like, okay, I might have a PTRS agent, which is looking at mechanisms of action, patient populations related, existing approvals.

00:20:48.380 --> 00:20:53.117
<v Greg Lever>I may also have data retrieval agents looking across scientific literature or conference abstracts.

00:20:54.019 --> 00:21:07.898
<v Greg Lever>I may have trial search agents looking across existing clinical trials, and so this then gets all brought back together by an orchestrator agent that not only brings in the insights from these specialized agents, but.

00:21:08.133 --> 00:21:15.473
<v Greg Lever>Can really understand the intent of my user's question, which is basically saying what's the potential of my early stage, asset?

00:21:15.740 --> 00:21:30.721
<v Greg Lever>And then by extension we can utilize existing molecular design approaches or target design approaches to say, and is there some better design that I could think of here, um, that will like really accelerate my clinical development?

00:21:31.890 --> 00:21:38.963
<v Selina Koch>So in protein design seems to be a little further ahead of say small molecule design, unless I've misunderstood.

00:21:39.865 --> 00:21:40.731
<v Selina Koch>Um, but there.

00:21:41.700 --> 00:21:57.281
<v Selina Koch>This neat idea I heard recently of, so if language models, they tokenize language, as you were saying earlier, where the smallest unit of meaning is the word as opposed to the way we learn language is building it up from letters.

00:21:57.749 --> 00:22:05.923
<v Selina Koch>that there is an analogy in the small molecule space of tokens or bits of molecules with certain functions and that you might be able to have like.

00:22:06.357 --> 00:22:12.864
<v Selina Koch>A tokenized library of functional components that could be built into anyway.

00:22:13.298 --> 00:22:17.001
<v Selina Koch>What, tell U.S. a little bit on the small molecule front, what you're hearing.

00:22:17.001 --> 00:22:18.537
<v Selina Koch>That could be a be a step change?

00:22:19.438 --> 00:22:35.953
<v Greg Lever>So I think what we can think of in terms of the protein and antibody modeling space is things like protein language models where that fundamental token or fragment is either an amino acid, if we're thinking about proteins or nucleic acids.

00:22:35.953 --> 00:22:38.723
<v Greg Lever>If we're thinking about other structures like DNA and otherwise.

00:22:39.124 --> 00:22:40.491
<v Greg Lever>And I think really.

00:22:40.858 --> 00:22:46.330
<v Greg Lever>You can think about molecular language models that actually build up this concept.

00:22:46.664 --> 00:22:51.403
<v Greg Lever>at the atomic level, and potentially also at the, the electron level as well.

00:22:51.403 --> 00:23:04.816
<v Greg Lever>And that's actually where a lot of my,  academic, studies and academic research was in, is in how do we build up a sufficient electron density to understand how these small molecules can best interact.

00:23:05.384 --> 00:23:08.720
<v Greg Lever>Um, with their targets and being impactful in a clinical setting.

00:23:08.987 --> 00:23:14.925
<v Greg Lever>And it's something that the language models are definitely going to, to really accelerate, um, in the coming years.

00:23:14.925 --> 00:23:17.162
<v Greg Lever>And it's a, it's a space that we really need to keep an eye on.

00:23:17.996 --> 00:23:18.930
<v Selina Koch>That was very cool.

00:23:18.963 --> 00:23:23.701
<v Selina Koch>And then if we wanna even go further afield, I guess in our imaginations here, what can happen?

00:23:24.102 --> 00:23:37.548
<v Selina Koch>I just heard a really interesting talk at our Grand Rounds conference on, um, the possibility of quantum computing, really speeding up simulations, particularly around all the electronic states and things like you were just talking about.

00:23:38.016 --> 00:23:39.584
<v Selina Koch>but that's, not here yet.

00:23:40.018 --> 00:23:42.186
<v Greg Lever>And that's probably also an entire podcast

00:23:42.421 --> 00:23:42.820
<v Selina Koch>Yeah.

00:23:42.953 --> 00:23:43.221
<v Selina Koch>Yeah.

00:23:43.255 --> 00:23:43.721
<v Selina Koch>Yeah.

00:23:44.189 --> 00:23:53.765
<v Lauren Martz>I think it would be great if we could get into a little bit about the clinical trial design AI applications, where we are now, and where we could be, you know, in the future?

00:23:55.133 --> 00:23:59.570
<v Greg Lever>Yeah, that, that's a really interesting point and I think there's a lot to to speak about, but I think what I would say is.

00:24:00.172 --> 00:24:02.606
<v Greg Lever>AI is really transforming trial design.

00:24:02.606 --> 00:24:07.746
<v Greg Lever>It's, it's doing things like quantifying site and patient burden.

00:24:08.113 --> 00:24:14.618
<v Greg Lever>being able to look at things like protocol complexity, but also simulating operational outcomes.

00:24:14.618 --> 00:24:27.665
<v Greg Lever>So, not only can we use these advanced analytics and other conventional machine learning approaches to really predict how different decisions within the design process, like eligibility criteria and visit schedules.

00:24:27.965 --> 00:24:33.438
<v Greg Lever>How they might impact recruitment rates, um, patient retention, um, or site performance.

00:24:33.438 --> 00:24:41.145
<v Greg Lever>But now with, increasingly generative ai, it's enabling these sponsors to, to really optimize these protocols, really say.

00:24:42.413 --> 00:24:46.750
<v Greg Lever>Okay, this is what may happen in the current state, but how do I make this even better?

00:24:46.750 --> 00:24:49.086
<v Greg Lever>How do I forecast that site activation?

00:24:49.354 --> 00:24:53.357
<v Greg Lever>How do I select sites with the highest likelihood of success?

00:24:53.357 --> 00:24:59.163
<v Greg Lever>And so really it's, it's getting to those pain points of, we know there's a lot of protocol complexity.

00:24:59.163 --> 00:25:00.365
<v Greg Lever>This is slowing recruitment.

00:25:00.365 --> 00:25:03.801
<v Greg Lever>We know there's uncertainty in these operational outcomes.

00:25:04.001 --> 00:25:07.038
<v Greg Lever>and also we know that the biotechs have these resource constraints.

00:25:07.372 --> 00:25:14.679
<v Greg Lever>And so really what you can do is you can, you can begin to demonstrate this value and sort of leapfrog across a lot of these constraints, by utilizing ai.

00:25:16.080 --> 00:25:22.586
<v Selina Koch>So looking across these different domains, target discovery, molecule discovery, clinical trial prediction design.

00:25:22.854 --> 00:25:33.932
<v Selina Koch>When you're advising biotechs today, or if you were to, um, and you had to give just a very short list of like, these are the most useful applications right now across those domains, like what would you say.

00:25:34.932 --> 00:25:45.343
<v Greg Lever>I think it's about making sure that you've, you've tried to have that coverage across the clinical development life cycle or the, or the pieces that are important for you.

00:25:45.609 --> 00:25:53.585
<v Greg Lever>to make sure that, okay, in the discovery phase, maybe you are thinking more about target id, generative approaches for design.

00:25:53.817 --> 00:26:00.057
<v Greg Lever>when you do get that, asset into the clinic, really how are you optimizing, your protocol?

00:26:00.258 --> 00:26:03.561
<v Greg Lever>And then also maybe when you are thinking, about the regulatory space.

00:26:03.929 --> 00:26:11.836
<v Greg Lever>how are you utilizing AI to really think about what are the compliance changes that are happening globally or in my target market?

00:26:11.836 --> 00:26:17.808
<v Greg Lever>And how do I need to basically de-risk, my development and be able to anticipate these things and act earlier?

00:26:17.842 --> 00:26:22.446
<v Greg Lever>And that's, that's in a nutshell what these AI approaches are providing you the capability of.

00:26:23.315 --> 00:26:28.452
<v Selina Koch>Is there any application right now where you'd say, just steer clear of that one, or don't expect a lot from that just yet?

00:26:30.422 --> 00:26:46.003
<v Greg Lever>So I think one of the biggest challenges that we might want to leave to slightly more, well-funded, outfits is the notion of in silico clinical trial simulation, and this is an aspect that has had lots of attention over.

00:26:46.304 --> 00:26:49.074
<v Greg Lever>Many years from many different types of approaches.

00:26:49.106 --> 00:26:58.016
<v Greg Lever>It's something that, generative ai, agentic ai, and specifically foundation models are going to have a massive impact in.

00:26:58.249 --> 00:26:59.584
<v Greg Lever>But it's still very early days.

00:27:00.484 --> 00:27:09.594
<v Jeff Cranmer>well, we've been talking AI in biotech with Greg Lever of IQVIA and Greg, I'd just like to get your closing thoughts?

00:27:10.761 --> 00:27:11.663
<v Greg Lever>Yeah, absolutely.

00:27:11.962 --> 00:27:15.834
<v Greg Lever>I think really one of the things to think about is that for those.

00:27:16.968 --> 00:27:20.270
<v Greg Lever>Biotechs, early stage biotechs, those sort of outfits.

00:27:20.571 --> 00:27:26.978
<v Greg Lever>agentic AI really offers the ability to compete at scale.

00:27:27.244 --> 00:27:54.172
<v Greg Lever>and whether it's identifying novel opportunities, designing smarter trials, you know, accelerating development with, with fewer resources, Not only that, the other approaches that we talked about, be it generative ai and also conventional machine learning, could be helpful in, in attracting investment, you know, utilizing these approaches to give this more objective, data-driven view of asset risk, but also asset value.

00:27:54.405 --> 00:28:00.077
<v Greg Lever>And so supporting things like capital allocation, the potential to de-risk these investment decisions.

00:28:00.278 --> 00:28:01.813
<v Greg Lever>I think that's gonna be really impactful.

00:28:02.614 --> 00:28:02.814
<v Jeff Cranmer>Excellent.

00:28:02.814 --> 00:28:05.150
<v Jeff Cranmer>Well, Greg, thank you so much for joining us.

00:28:05.182 --> 00:28:09.220
<v Jeff Cranmer>the way things are moving so quickly, we're gonna have to have you, uh, back on tomorrow.

00:28:09.220 --> 00:28:10.055
<v Jeff Cranmer>I don't, I don't know.

00:28:10.055 --> 00:28:17.695
<v Jeff Cranmer>It's, uh, tough to keep up with everything that's going on in, uh, how AI is helping to improve, how we work in biotech.

00:28:17.895 --> 00:28:26.938
<v Jeff Cranmer>so once again, this has been, uh, the BioCentury analyst team, speaking with Greg Lever, Director of AI Solutions Delivery from IQVIA.

00:28:27.337 --> 00:28:44.588
<v Jeff Cranmer>a special thanks to IQVIA Biotech our sponsor, as well as Kendall Square Orchestra, the Boston based, ensemble that, does the music for all of BioCentury's podcast Tickets on sale now for their new season.

00:28:44.855 --> 00:28:49.894
<v Jeff Cranmer>we will catch you on Monday with the regular BioCentury This Week podcast.

00:28:51.162 --> 00:28:56.000
<v Alanna>BioCentury would like to thank IQVIA Biotech for supporting the BioCentury This Week podcast.

00:28:56.300 --> 00:29:04.709
<v Alanna>To learn more about how IQVIA Biotech can help you turn your vision into venture capital, go to IQVIABiotech.com/visionaries