Improving Drug Development and Decision-Making with AI

The Life Sciences DNA podcast is sponsored by Agilisium Labs, a collaborative space where Agilisium works with its clients ranging from early stage biotechs to pharmaceutical giants to co -develop and incubate POCs, products and solutions that improve patient outcomes and accelerate the development of therapies to the market. To learn how AgilisiumLabs can use the power of its generative AI for life sciences analytics,

To help you turn your visionary ideas intorealities, visit them at labs.agilisium .com.

You're tuned to Life Sciences DNA with Dr.Amar Drawid.

Amar, we've got Dimitrios  Skaltsas on the show today. Who is Dimitrios ? Dimitrios  is the CEO and co -founder of Intelligencia AI. That's a provider of AI -powered solutions that support data-driven decision-making in drug development. Before launching the company, he spent more than a decade in various healthcare and life science -related leadership roles. He led McKinsey's new ventures for big data and AI in the pharmaceutical R &D sector where he pioneered machine learning enabled enterprise products in life sciences. Prior to entering the health tech industry, he began his career as an at torney. He holds his MBA from INSEAD and a law degree from the University College of London and the University of Athens. What's Intelligentsia AI seeking to do? Intelligentsia AI is helping biopharmaceutical companies accelerate and de-risk clinical development by applying the latest AI -based innovations and is built on a foundation of expertly curated and harmonized clinical and biological data. So by having more accurate and AI-driven predictions around a drug receiving regulatory approval earlier in the process, companies can make better decisions when it comes to their portfolio of assets. Now, what are you hoping to hear from him today? I'm hoping to hear about Intelligentsia AI's unique approach to calculate which clinical trials will be better successful and how that is driving biopharmaceutical companies' decisions and how it's raising the bar in the space. Well, before we start, I just want to remind listeners that if they would like to keep up on the latest episodes of Life Sciences DNA podcast, they should hit the subscribe button. If you enjoy the content, hit the like and let us know your thoughts in the comments section.

If you want to listen to this while you're on the move an audio only version is also available on major podcast platforms. With that, let's welcome Dimitrios  to the show.

So, Dimitrios , for joining us today. We're going to talk today about how biopharmaceutical companies make pipeline decisions about clinical trials and the potential for AI to enable faster and better decision -making about what programs to advance. So let's start with the problem the industry faces today. So it's well known that drug development is very expensive, time -sensitive, time -intensive, and with a lot of failure. So can you...

Tell us more about this. Can you please describe the problem in more detail for us? Thank you for having me. The industry inherently is risky. At the same time, the industry is ready for a breakthrough. So if we think about how it works, to bring a new therapy to the market, to the patients, you have to test on on Why? Because water whatever on animals, pre clinical

it doesn't necessarily translate well into humans, right? And for all good reasons, a very highly regulated space. Now, aspart of this, as you mentioned, you have high costs. So it's estimated that, you know, a a therapy costs about in the hearts hundreds millions or sometimes in the billions. It takes a lot of time, on average, say 10 years. And what is very crucial here is that...

this process is highly risky. So, 85 to 90% of the programs eventually never make it to the patient. Part of this risk is inherent to the process, as we discussed. But part of this actually could be better understood to and mitigated. And that's where intelligence comes into the picture. Okay. So, see, the thing is like, can you talk about like, what are some of the biggest problems

in the success of clinical trials. So 1 you mentioned, which is, yes, we work with the animal models and then the drugs go in there and animals are not human. I mean, the lab animals, their biology is different from that of the human. So we don't know how that success is going to translate there, right? But apart from that, what are some of the other problems you see why clinical trials fail?

So what we have seen again and again withour partners, with our users in the industry is that when you think about portfolio strategy, portfolio management, this type of decision making, often you lack workflows or you have protein workflows. There are a lot of biases, human biases that are crippling into decision making, lack of consistency across different groups, different people,

and lack of consistency. So we've run a survey with people in portfolio and project management with over 100 participants. And what we've seen is that about 6 % they see as the highest challenge in the space

the lack of consistency. Now, a significant number of, again, same people, about 20, 25%, they already use some AI. And that's an important shift from maybe a couple of years back. That's also an indicator of how much more need and opportunity there is in the space to actually leverage more

standardized approaches and more accurate approaches. Okay, so when you say consistency, like that's basically consistency in decision making, which is that the same criteria do not like for starting the clinical trials, they don't get applied consistently for everything. Like is that accurate or what else? I'm taking maybe a more specific view when it comes to portfolio strategy. So if you think about phase1, right,

I mean, a major pharmaceutical company would have a dozen or dozens of candidates in phase 1. It's very hard to apply some objective metrics because you don't have efficacy, you don't have safety signals. And often we're talking about novel therapies. So, the biology is not fully proven, it's not fully understood. So, what happens

is that people apply some benchmarks and baseline. Most of the times a very cheap news quality. People complain about this and people will discuss with key opinion leaders in a process typically called expert elicitation. And you'd come up with a semi -quantifiable approach. Now this approach is not necessarily consistent from one programmanager to the other, and the human...the human intuition there. There is anincentive at the program level to, and not only an incentive, there is a general interest in actually promoting the program you're working on. And at the portfolio level, it's very hard to qualify actually if you're comparing apples with apples.

Yeah, and it's like it's someone's baby, right? Like we're like driving the clinical trial and then they want to they have their personal interest in really driving that forward, no matter whether the trial is being successful or not. As it should be. The question is how we equip those people with better tools. Both at the program level and then at the portfolio level. Yeah, yeah. You know, everybody speaks the same language. Everybody applies the same.

the same record in metrics. Yeah. And what I've seen is like in terms like this probability of success, right? Like you have for phase 1, certain probability of success, but that's like a big range. I mean, it could be better from 50 % or like, you know, from 20 % to 60%. And then at least like when I've, you know, I've been involved in this kind of like decision making has been, okay, well, yeah, yeah, it should be like for this therapeutic area, it should be maybe 30%, but

we really have a lot of confidence in our drugs. So we're gonna make it 40 % instead of 30%. I mean, that's like, there's no data. It's just, okay, well, I just feel that way, right? And so I'm gonna put it more. And then also about, right? And then you're also like, there's also this net present value tied to this, right? So if you increase the probability of success, then the net present value becomes positive instead of negative. And then you're...

So like a lot of these kind of, I've seen like, you know, my my people just putting these numbers on just to get the right outcome rather than that being based on a lot of science or a lot of the historical data. I mean, you've been at some of the largest track tracking organizations out there. I'm sure you've seen it. And as I have seen the cross now, multi-partners.

It tends to be the norm, it's not the exception. Large organizations will have often quite sophisticated people and modeling for late stage, phase 3 that will apply Bayesian techniques and so on and so forth. But there is typically lack of consistency, lack of certain sophistication, especially in the earlier stages. And it's unfortunate.

Exactly. Because as you said, this, the risk, again PTR as a probability of success feeds ultimately into a broader EMPV calculation. It's, you know, along with commercial success, along with the revenue estimate, it's the hardest thing to get right. And the industry does not necessarily have the appropriate tools to make this important decision. Which ultimately drives...so many things in decision making in pharma. What to prioritize, what to expedite, what not moving, gonna go decisions, investing in these major phase 3 trial, ED &L in licensing assets, acquiring as companies. It's inherently risky business.

That's why there are the incentives by the governments for all good reasons and purposes. And what we aim to do is address this question that goes at the core. What is the risk? And this is in aquantifiable, consistent, unbiased way. Yeah. So see, like we talked about probably of success or POS. You also referred to PTR as the

probability of therapeutic and regulatory success. So for the audience members who are not very familiar with this concept, can you describe that a bit more? Can you break that down for us about how the probability of success works or how the PTRS success works? It's matter of definition, as always. I'll start with them. What we define in the industry as success in the PTRS

is the regulatory regulatory down the line, right? And that goes a long way if you are able with more accuracy to predict a phase one. 1. At least what may happen seven, eight, 10 years from now, that's major. Another type of success is transition probability. So the probability of transitioning,

program from one stage, from one phase to the next. And that's also very critical because that ties to the go -or -not-go decisions. So we do both of those and we work with our partners in both of those. There's also typically when you think about probability of technical success, it's more from one phase to the other. Probability of regulatory success is ultimately, you know, when you have filed with the regulator.

What are the likelihoods? What are the chances? And then the compound metric probability of taking an end -regulator success is, you know, dependent on where I stand in clinical development, phase1, phase 2, or phase 3, what are the chances that ultimately I will get a regulatory report? Okay. And so as, so, and you're then, let's say something is like in the phase 1 or so, then you have to multiply out

the probabilities of successes of all these, like, you know, the technical success in phase 1, then the transition probability to phase 2, then again, technical success of that phase 2, then transition to phase 3, technical success of phase 3, and then regulatory success, right? So all of that, the multiplication. So that's why, like, I guess when you start the clinical trials, the probability of success is very low usually, right? So you're exactly right in the way it's described then.

Yes, it tends to be quite low, especially for people who are not in the space, sometimes it's so too low. You know, inoncology, for instance, it's about 5%. One in 20 programs will eventually succeed. Now, a drug may have multiple programs. So, this doesn't translate necessarily to success of the drug contents, but translates into the programs where actually you allocate also the resources.

In other therapeutic areas, it tends to be often higher. It can be 15 or 20 percent. So it varies. And also with inoncology, for instance, different diseases, different indications tend to have also different success rates. Okay. So can you tell us more about like what are some of the therapeutic areas or diseases where there is like higher probability of success? What are some of the

therapeutic of areas and diseases where there's lower probability success that you've seen like historically? It depends on if we take the extremes, you have the Alzheimer's of the world. Success rate is very low. Why? Because partially, you know, the biology is not fully understood. You have other diseases where you have a successful compound and then...

And then, you know, for instance, inoncology, you create combinational therapies. Yeah. And they tend to be better sometimes even better than the mono therapy. And cumulatively, the probability of success in this indication goes up. You have other cases where, you know, there's some operational risk or everyone goes after the same patient pool. It's hard to recruit patients

and trials for [undecipherable] access down. Wow. There's so much competition. So it varies rise a lot. Pancreatic cancer is also another in famous. Sorry, which one? Pancreatic cancer. And that's important also

to have always in mind when people think about probability of success and making investment decisions, making resource allocations, and so on and so forth, a 10 % is not always same as another 10%. So if there's a 10 % in pancreatic cancer, for instance, on a phase one 1 and a 10 % in...

non -small cell lung cancer. These are not equal because in pancreatic cancers there is higher unmet needs. So maybe ultimately both programs have the same chances of getting regulatory approval down the line. But potentially a company would be more willing to take the bet on the pancreatic cancer. Yes, yes. Because I mean, it's a, I mean, I remember like it's about,

even though the incidence of pancreatic cancer is much lower than that of lung cancer. So a lot fewer patients get it, but then still in terms of the mortality, almost about the same number of people die from pancreatic cancer as lung cancer. I mean, it's staggering how many people die of pancreatic cancer. Yeah. And it's also interesting, Dimitrios, what you mentioned about this combination therapy

having probably like a larger probabilityof success. I'd never thought of that before. Like you have your new drug, which is you're combining with another drug that's already there. I mean, yes, it has a higher probability of success, right? So, but again, it depends on what you're comparing against, right? That's also the thing. Exactly. Let's not generalize this. Okay. What we have seen also, you know, it depends on these indications. So there are indications where...

You know, combo therapies tend to have higher success rates than monotherapies. And there are other indications where it's exactly the opposite. Yeah. Okay. And that's a bit of the beauty of our space, of our industry, and also of thinking about risk, probability of success. You know, it's, you have to qualify things. It's complicated and you have to couple eventually whatever the...

AI in our case indicates with human intuition. What we built is more of a support mechanism. Why? Because it's such a complicated question and humans have to come in and qualify and complement things with their own experience, with their own intuition. Yeah. So now that we've talked about the problem here,

so what can the application of AI do to address this? Right, so as a company, we focus mostly on applying AI, machine learning, on assessing and qualifying the probability of success, right? We do more things, but that's the core. And we have a patent, we have IP patents, right? And we're the first company actually in the space, but in 2017 wow.

to take a bet if you will that AI couldactually yield some strong insights and results here. Now, if you think about user case and user personas, right? Typically we work with R &D and with DNS, there's an evaluation, right, or corporate development. So it's people who...

qualify either at the portfolio level or at the asset level. And people also doing license or acquiring assets. So if you're a large pharma, again, you want to level set the risk assessments across your portfolio and make decisions accordingly. If you're a program manager, you know, you're interested in lifecycle management, how you...

actually drive more value out of your asset and can risk and for this is like going to go like indication selection and so forth. If you are in a certain value or in an EDNL group you want to identify earlier than others what may be a hidden gem you may want to qualify actually what is the risk in this ENPV calculation

Yeah, for license or something.

These are the main use cases. Okay, okay. And see, AI, like there's no shortage of places where AI can be applied to improve decision making. So why did you choose, you know, start with this drug development and why do you think this is AI is particularly well suited for this?

You know, I worked for, I've been in this space for a bit over 15 years now, and I worked for several years as a consultant. And I was with McKinsey doing a lot of work in utilities and M&A and commercial strategy, portfolio strategy. And every time that an NPV calculation is involved, you have to assess probability of success.

And most of the times it was an underwhelming experience, like people moving and literally they shoot in the dark because they don't have the appropriate tools. They don't necessarily have the support to do any better. So I had this experience and then at some point we're experimenting with big data and AI and there was this uniform crew, new ventures where I came in and led the efforts

in the R &D space, pharmaceutical R&D, both discovery and development. And at the time, we're building very interesting solutions in both discovery and development. There is an eventual Intelligencia focused on this problem, and I focus on this problem, is because of its nature. It's strategic. There is a huge unmet need there. You see the AI track discovery space, and there are so many wonderful companies

doing great things. And I was strikingalready back then. But when you think about risk, no one had ever tried to apply AI. And that was a bridge to Intelligencia. We decided to take a bet that no one, including McKinsey at the time, was willing to take.

So yeah, it's helping all these people out there to track development with better tools so that they can make better decisions. It's a major problem. It goes into the heart of track development. So Intelligencia AI has developed a portfolio optimizer. So can you talk more about what that is and how that is used?

Right, so the name of our core platform is Portfolio Optimizer. What tracking does is it brings together our insights onprobability of success into a software interface. Why a software interface? Because we think that's important for users to have direct access to data and insights and important to be able to work

in a self -service mode as synchronous. Plus it's more scalable. So that's our Portfolio Optimizer. Core functionalities that are unique to it is that beyond our probability of success, which is industry -leading, I think in many accounts, it provides also explainability. So it moves away from the notion of having a black box.

Okay. Which scares a lot of people and also is limiting in its users. And we provide visibility into the drivers of the machine learning model we have developed. And what is important, by how much, so that we can help with decision making in a more meaningful way. It goes beyond the portfolio strategy and that's also an asset strategy. How can I improve

my decision making and my probability of success also at the program. Beyond that, we work also in other ways. We have an insights excellence team. These days, it has always been about specialization. It's really about insights, how you should inform decision making. So we have a specialized team that works with our partners, with our users on very specific, at -home questions that will make it

be the sensitivity analysis, be going deeper into the data, and that's a special question. We provide also access sometimes to our data later. People have come to appreciate a lot the comprehensiveness and the quality, the accuracy of our data. And surprisingly, to me, initially,they're used in much broader use cases. A lot of users in competitive intelligence, a lot of users in trial design.

So yeah, it's a matter of things. Okay. So you talked about explainability. So when you say that, you mean that like whatever you're saying in terms of the probability of success, your algorithm will explain why that is the case? We back up. So the algorithm is being fair with...

data in a structural way. So pretty much we do feature engineering where human intuition comes in a major way. And yeah, we provide the parameters to the users. So the fact that, I'll take an example, you know, biomarkers are being kind of...

included in this trial, trying for legal success up or down by X percent. And there are some, what I always discuss with our users, there are inherent signals and there are external signals. It's important to take into account both. Inherent signals could be the efficacy, right? Once you have some efficacy, then it's easier to directionally...

and if you want the probability of success.There are also external signals, for instance, the fact that you may be recruiting more patients than you were planning initially. It's something that, it means that you know something as a user that you may not have published to the world, but we're picking it up. So for accuracy, it's important. It may be less actionable

than a feature about which endpoints are being used. A lot of phase 3. But it's still very important for the accuracy of the ultimate prediction. Now, I mean, see, the interesting thing about AI is that you have to have really good quality data and a lot of data. So can you tell us about what

What is the range of like data and the sources of data that are driving this? Yeah, I'm glad to. And again, we have IPover here, so it's protected and it's out there in the world to some extent. So yeah, I mean, whatever you can share publicly, right? It's all about data, at the end of the day, or it's most about data. A lot of talk about AI, even the chat GPT models these days.

I know so many people complain about hallucinations and so forth. Why? It largely depends on not only how you train the model, but what data you feed the model with. So in Intelligencia, we have purpose -built data to feed models, and we have invested a lot into our datafrom day one. I'll discuss what data we have and why our data are better than...

other data. I'll start with the latter. So we have harmonized data, which is very important. There's a lot of data out there, but they don't speak with each other. And so what we've done is we've defined ontologies that help us pretty much bring dictionaries that help us bring data together in a harmonized way. We don't have a lot of automated data pipelines that help ensure...

timeliness and then we have human interloop. We have experts who add curation layers to the QCs that we already do in an automated way. That ensures quality. Now, that goes, all this goes along way and it's hard to get right. So that's the foundation of this. Now, what type of data?

biology, drug and disease biology, clinical outcomes, if any. Sometimes they're available, sometimes not. But efficacy and safety in a very pain stakingly detailed way. Clinical trial design, setup, execution, patient populations, which end points, and so on and so forth.

And lastly, regulatory data. It's a highly regulated space. There are signals in what regulators are saying in regulatory paths and so forth. Okay. And what's been done to train and validate the portfolio optimizers so you can predict this probability of success with very high accuracy?

So we claim we have the highest accuracy inthe space. I would say it's appropriate also to claim that as far as I know, we're the only ones who have been prospectively validating our results in a statistically significant way. wow. In a repeatable way. So as examples, and we published recently an article in

Fierce Pharma that's citing some these. Weprovide some predictions in the earlier days of the company back in late 2019on NSCLC and melanoma working with a large pharma. We have been tracking nowthose monitoring those decisions. You know they have been tracking out quitewell over time.

They haven't proved our predictions. Our current predictions are very sometimes significant. But even then, we're able to predict novel therapies quite likely. And that was for early stage phases, phase 1 and phase 2. That's one example. What is a more interesting example to me is the following. Again, discussing this article. There's a thesis that the main arbitrage, when you think about investing in a biotech company, is risk,

right? And the market value, the stockprice is driven mostly by how investors, how the market understand risk. So what we've done is we said, okay, we have an Intelligencia Services company score based on risk. What if we force rank the companies, early stage biotech, publicly traded? And what if we, you know,

track, monitor how the stock market moves for these companies. We have done extensive back testing on this. Last year we decided to do prospective validation. So April 23 we said, you know, we look at companies with a market cap of 100 million and 1 million. They were in oncology and I &I, immunology, inflammation. And let's pick the top.

top, I think 20%. Right. So what we saw a year later and published on this is that, you know, these top companies, they had the market returns of 6 % on aggregate, whereas the remaining companies in that basket, they have, and the basket was about 150 companies. The remaining companies had about 15%. So,

the high companies were beating them and also the ETF in the space, XBI, are also about 15 to 20 percent. So that's promising. What we also observed is that out of the top 10 companies in our portfolio, the top three of them got acquired recently. In the meantime, we have identified those companies early on and we have identified them with what I call the most sophisticated

unsophisticated way, right? If you're a practitioner, if you're an investor in a hedge fund, if you're a practitioner in BDNLs, there's some value there. You're very sophisticated and you look at an array of things as you should, right? What we, we don't do this, but what we do extremely well is assessing risk. And because risk is such an important parameter in ENPV, you know, it's a very nice alternative

complementary approach to identifying opportunities. And we do increasingly work in that space, including the financial market, like research analysts, hedge funds. It's a very interesting use case. And to your original question,

there's a lot of AI out there. The important thing is how well -validated it is. And it no longer cuts the mark when people come and say, yes, we have an 80 % AUC and validated rates respectively in this test set. It's a good starting point. It's a necessary starting point. But what we have observed in the industry with our partners is that ultimately people ask for...

actual real -life publications. That's very interesting and I can see the big use case with these really important financial decisions, where all of this information in terms of BDNL, business development licensing, that's where this can play really a huge role. For those in the audience not familiar with this net present value, this is where basically

we're doing this calculation about what is going to be the cost, what is going to be the revenue, and what is going to be the success of getting that revenue, right? And so, you're putting all of that together. If the net present value is positive or it's like very highly positive, that's when you're making the decision to do the trial or to buy the company that has the product, whereas if it's negative, then you're saying, no, this is not a good business decision. So,

having the right inputs into this model is so critical to make these large financial decisions. Yeah. And one question I have here is that, see, there has been a lot of talk about low probability of success or so. In your experience, have you seen clinical trials getting better

probability of success over let's say last couple of years compared to 10 years ago. Have you kind of seen the shift in pharma or not really? Is that just talk still?

In my experience on aggregate, I think it does not apply. We have the ability to analyze for real -time success by timing. So we track that. It's an interesting exercise, but in my experience, it may apply in some cases, but in many others, it doesn't apply. What has been my experience though is that...

Exactly, because there's a declining R&D productivity in this space. It's a burning platform. It has been declining for the last many, many years. There's an increased emphasis and attention internally, especially in large pharma on, but also mid-cap and open biotech, on how to make better decisions, how to improve, for instance, the success rate from phase 2,

phase 3, from phase 3 to approval and so on and so forth. Sometimes earlier, it depends on the company. There's an increased emphasis. I think this will ultimately lead to an improved overall clinical development success rates, which can be huge for the patients primarily and for the industry, which still spends hundreds of billions of a year in R&D and hundreds of billions in DTNA.

And the answer ultimately is better decision making, which will be a combination in my experience already with large pharma, but also mid cap and biotech. But mostly with large pharma is a combination of all the trade investments they do internally, on better data, better workflows, better data science. Some of the pharmaceuticals out there do amazing things. Coupled also with triangulated approach with specialized decision support systems, such as Intelligencia. Not the only one. I mean, across the value chain, there are many, many of those, I call them gem companies, you know, the specialized or something that they do really well. They have outsized them. Okay. Now, now,

Intelligencia AI operates with this software as a service model. So who in a biopharmaceutical company would use Portfolio Optimizer and do they need any specialized training to do that?

Yeah, it's portfolio strategy teams, it's program management teams, it's search and evaluate, PTNL. and NL. We see a lot of CI, increasingly competitive intelligence. Do they need specialized training? We always make sure we provide some out -holding, appropriate onboarding. It's needed. And again, it's largely a technical platform.

We've gone a long way to have a user-friendly platform, and we get a lot of compliments for that. But yeah, we do provide onboarding, we do provide hand -holding on questions. And ultimately,there are users who get it right away, and there are users who may go deeper in some specifics of the functionalities of the platform, or they...

and maybe more inquisitive on certainthings and we provide that support. I thought maybe maybe could talk about the work you did with the ZS Associates as an example of how this can be applied in the real world. So can you talk about what ZS was trying to do and how they use your technology and what the result was for that?

Yeah, I'll take it as a broader question of, you know, give us specific use cases, right? With ZS we have a partnership.

You know, sometimes we do work together and quite often we don't, right? But it's been a great partnership over the last three years. Expect to continue for a longer time, for another three years plus. Now, an example that I always like to highlight when it comes to using it is the following.

In a course -packed discussion we had about workflows and consistency and so on and and forth. We hadn't worked enough now a couple of years with a large pharma. They trusted us, they started using us, but initially we were not part of the workflows. It was like something optional, complimentary. And how this phase 3 trial where they were already spending a lot of money enrolling patients and so forth, they got the interim results and they were not good.

And what they observed is that internally they had assigned a probability of success of over 40 % where Intelligencia AI were providing about 10%, which is very low for phase 3, by the way, significantly lower than the historical average. So that actually led to a workflow update where now they have to triangulate

when there's a discrepancy, when things are consistent, it's fine. When there's a discrepancy, there's internal dialogue. It instigates internal discussion. And that has been very helpful to this organization because it doesn't really add any burden. It only makes them smarter in the case where they have to press these things more. Sometimes they learn things, sometimes...

Sometimes they drive a different decision making, sometimes they don't, but it has been enriching their process and making them smarter. Great. Now, the company also has this another product, the Clinical Development Insights that's in beta. Can you tell us a bit more about that?

Absolutely. I'll approach it in two directions. The one is, Portfolio Optimizer is a very advanced AI platform. Interms of insights so far, it's mostly descriptive analytics. There is a beauty in simplicity. And if you think about it, most people in this space, actually what they lack is not the over -sophisticated system. Often they lack basic systems. In this case...

one component in the clinical development insights module is highly curated data where we provide insights, we analyze and provide insights. And over time we'll have multiple modules. The first module is the good old school baseline on historical likelihood of approval, which unfortunately the industry...gets

suboptimal support right now and there's complaints about that. So if you want to have a baseline for the likelihood of success for let's say NSCLC with some specific parameters and let's say primary deployment there's no solution right now for this. There are no benchmarks that are granular enough,

are no best ones that have good enough code of data. That's the gap that we're filling here. And we go a step further where we show, you know, path to path analysis. You know, when you want to design your...

create your development plan and say, okay, I'm moving from phase 1 to phase 3. It happens sometimes. I'm moving right into phase 2. And skipping phase 1. And so on and so forth. You know, we provide insights into how those programs, how those paths have served over time. So we share with decision making there. Over time, we'll be augmenting these with insights into...

clinical outcomes and trial design and endpoints and all those crucial parameters for appropriate clinical development decision making. You know, the aim is to support a broader group of people, including people who do the trial design. And as I mentioned, we're doing more and more work with complete intelligence. So that's...

That's what I'm pretty sure. Well, you talked about today, right? This is getting to really the crux of drug development, right? And if we can have better decisions around what clinical trials to do, and set up the clinical trials in the right way, that will help tremendously in terms of getting the right drugs out for the patients

in a much faster and more efficient manner. And I think that would be amazing to have that. So I really appreciate that. Dimitrios Skaltsas, co -founder and CEO of Intelligencia AI. Dimitrios , thank you for your time today.

Amir, thank you. Love discussing with you.Thank you for inviting me. Well, Amir, what did you think? Yeah, it's a fascinating area, the probability of success and calculating that and making decisions around that, right? So I definitely have thought that there is a bigun met need in this area because I mean, I've worked with many of my companies where we...

needed to make the decisions around the clinical trials and around the portfolio. And then a lot of times we just didn't have the right data. Well, as someone who's lived in the space, I've never met a pharmaceutical company that didn't describe itself as being science driven and data driven. It's still a human enterprise and other issues have away of coloring decision making.

What's the potential for AI to make decision making more rigorous? So in this aspect, right, when you think about the probability of success for a clinical trial, what do you base it on? Imean, you base it on a similar type in this specific therapeutic area, in this specific disease, you look at the historical information, right? Like what are the historical trials? What has been their success rate?

But then you're also looking at the biology of this. I mean, some points that Dimitrios  has brought up are like, are you going to get enough patients to recruit? Right? All of these, like there are a lot of, there's the scientific kind of questions. There are some of these more practical kind of questions. All of these are going into the kind of determination of what is the probability of success. And of course you don't know because human biology is so

difficult to really understand. We don'tunderstand it very well even now, although we have made tremendous progress in the last 70 years or so. But still, you don't know. And so when you think about it, well, the probability of success could be 20%. It could be 40%. It could be 50%. You really don't know. So there's a lot of lee way in terms of what is it that you want to put in your model if you're trying to see if you should do this clinical trial

and what the overall, in the end, whether this is going to be successful or not. Because you have such a large window of which you can put in the probability of success, it becomes quite a bit subjective rather than objective. That's why I believe that with AI, it will be very beneficial because AI is not going to have any… It's going to be looking at the data in a very…

non chalant manner and saying, okay, well, this is what it is. We should not be, like, there should not be like much about it. Okay, well, I just think this is a great compound. That's why we should do it. I mean, I've seen that a lot where people increase the probability of success because they think, yeah, yeah, yeah, that was historical, but I really think this is a fantastic compound. We should put more probability of success for that. But I'm pretty sure like for all the other trials that failed,

people said the same thing. I mean, people are doing the trial because they are so excited about this, right? So we need to take those kinds of irrationality out of the decision making here. And I think AI would be really good for that. Well, it does seem to be a place where AI could make a big impact. But as with other AI cases, it comes back to having the right data. It's not just internal data about a program, but also external data. Do drug makers have access to the right data to

de -risk these decisions? So the drug makers usually may not have so much access to all this. Also, this is very time consuming because in a particular disease here that you're looking at, you're looking at a bunch of, a large number of historical clinical trials, what exactly happened with them, what were the compounds.

And then your company, a lot of other companies, what happened in terms of regulation, how does the disease make them look like? So, for just one clinical trial to do so much data exploration and data analysis is not a very easy task. And so, what I like about what Intelligentsia AI is doing is that they're doing that across the board, but then…

all the data they're putting together, that can be actually applied by a lot of pharmaceutical companies who are in the same area. You can think about it could be even a pre -competitive space because you're basically looking at, okay, well, what happened? How's the biology around this? Well, maybe not about the specific clinical candidate, but other than that, what happened historically is the same across the board.

So there is a lot of benefit in terms of putting all that together for multiple companies and so just one single pharma company doing that for every single clinical trial is a lot of work and it's so I don't think people have the resources to actually do like very comprehensive analysis for this. If we do get to a point where AI is effective at informing these kinds of pipeline decisions is there some risk that we end up excluding or not pursuing novel therapies that don't fit the AI algorithm?

There definitely is a risk around that. But although I would say that you can maybe account for that, right? Like that novelty of this, and there's always a chance that something for which youdidn't have a whole lot of data and then you adjusted for that, right? But Ithink that risk is still there right now. I mean, I think right now, the way I've seen the decision -making around that, there's a long way that we can improve a lot on that. So I would say at this point,

I'm not worried about what the AI maybe missing, maybe like a trial there. I would say right now, if AI can make the decision making consistent, we will see really a big benefit to the patients because of that, right? Because then the right clinical trials are being selected and then they're running efficiently. And so we have a higher probability of getting the right drugs on the market for the patients earlier. So I would say I see much more benefit for AI right now

rather than the risk of using AI here. And also, as I said, there's explainability, right? So why was the probability of success 23 % and not 30 %? They're trying to explain that, right? And that's something that you as the human clinical decision maker, you can look at that and see if that makes sense or not, right? I mean, and then you can make your own decision about it. Well, it's a compelling application. Until next time.Thank you, Danny.

Thanks again to our sponsor, AgilisiumLabs. Life Sciences DNA is a bi-monthly podcast produced by the Levine Media Group with production support from Fullview Media. Be sure to follow us on your preferred podcast platform. Music for this podcast is provided courtesy of theJonah Levine Collective. We'd love to hear from you. Pop us a note atdannyatlevinemediagroup.com. For Life Sciences DNA and Dr. Amar Drawid,

I'm Daniel Levine. Thanks for joining us.