393  How AI Changed the Business of Drug Discovery
Dr. Michael Zaiac, Novartis of Europe

Dr. Kevin Folta, Talking Biotech Podcast
===

[00:00:00] Kevin Folta: And today our guest is Dr. Michael Zaiac. He's the head of Medical Affairs oncology at Novartis of Europe. So welcome to the talking Biotech, Dr. Zak. 

[00:00:10] Michael Zaiac: Hi Kevin. Thank you for having me on your podcast. 

[00:00:13] Kevin Folta: Yeah, you're very welcome. I think it's a really exciting addition to the series and really to learn more about how companies operate and, and how they adjust to changing times.

And let's start out just by talking about some of the basics is how is the integration of technology. Into pharmaceutical r and d and, and medical affairs really changed over time and, and what are some of the key milestones in that journey? 

[00:00:38] Michael Zaiac: Okay, great. That's a great question, Kevin. Um, I mean, I, I compare this, um, a little bit to my own journey with digital, uh, which started about 35, uh, 40 years ago, uh, in, uh, research and I think the first 10, 15 years of that journey.

Both my own as well as in corporate was of Plagiar. I think we copied a lot from what was available. We took a lot of auto automatization from automobile industry and others, and that was Incorporat. Into our production. Um, and I remember myself, I, my, my first research project some 35 years ago were with regards to, uh, human heart transplantation and bridging the gap with artificial hearts.

So we really had a lot of technology from Automo. And then I think, uh, accessible, workable PCs came into being. And I think what they did, um, they allowed us in, in medical science, um, to publish and to to write better and to really probably that. Spun a wave of, um, publications of, of making things available to the public and I think really helped medical sciences, um, from that aspect.

And I think the next phase I think of digital and corporate medical and incorporate pharmaceutical. Was dedicated systems based on that increased computing power. And a lot of these systems came in production of medicine. So in a white pill production, but also in biological production. Um, and then later on in research in robotic testing of, um, Our library of compounds against various targets.

Um, so basically taking away people, um, doing it, and by that, increasing the throughput, many, many, many faults allowing us to test many compounds rapidly in cell systems. Um, and then more at the customer facing end. I think the increased computing power and handheld computing really brought, um, customer relationship management systems.

Um, The ability of managing large sales forces and generating insight and customizing their behavior. That's sort of the more backend of digital, which, which came into pharma. Say something 15 years ago, and I think the last phase after this plagiar and then the more dedicated development phase, I think.

The intellectualization of digital, and that's for the last 10 years. Now we move beyond replacing manual labor, um, with computer and robots to replacing intellectual labor with a AI systems. And, and these systems, um, allow us to. Better predict targets to allow us to predict how the proteins we synthesize may fold in the RY state.

Guessing their biology much better. They allow us things like are attacking of our medicines. Um, ship. They allow us to better meet patient and customer needs by stop second guessing them or better second guessing them using artificial intelligence. And I think we will discuss many of these aspect in, in, in this podcast.

I think throughout these three phases of, um, digitalization in the pharmaceutical industry, there are. For me, two common icons. I think we were trying to be in sync with our customers. And the other one is that, um, this kind of technology change always needed a considerable amount of cultural change. So although I think we are always talking about, uh, being digitally adopted, the harsh reality is that our talk precedes our actions.

And I think we need to always be very careful when we. The next stage of digitalization that we move, the culture that we implement, change along with it. 

[00:05:59] Kevin Folta: Well, let's start out by talking about Novartis of Europe and mostly your area of the company. What exactly are you working on? And give me a good sense as to how that fits into the broader context.

Yeah, 

[00:06:13] Michael Zaiac: in, in my area, which is more sort of towards the customer facing and we are basically working on two aspects of digitalization, intellectual, digital use. And that's, um, using artificial intelligence to better identify patients, to really make sure that. The patients will get the medicine they need.

Not too much, not too little, not too aggressive treatment, not too conservative treatment to make sure they have the right effects of medicines. Another area we are using, Digital technology is, um, what I call the double personalization of information. So yes, in the past we could have had a call center where we tried to meet the doctor's needs for calling in, say, when they were on that night or weekend, if.

But now I think what we have is we can meet the doctor's demand better by forecasting it using the AI systems. And on top of this, we can access large databases and customize the information we give to the doctor to make it right for the patient. And I think that's what I call double personalization, and that's one of the area we are dealing with more on the customer facing side.

Of pharmaceutical industry in my teams. Okay. 

[00:07:52] Kevin Folta: So you talk about this idea of customization for the, for the customer, but what data are going into that? Is this genomic data or is this maybe some other kind of behavioral data? What data are going into that equation to give that prescription to mesh with what's happening with pharmaceuticals to guide physic.

[00:08:14] Michael Zaiac: It's a, it's a little bit of both. So to personalize it to the cu, the principle customer for us, which is the physician, it's of course behavioral data. Um, what is this physician normally asking for? When do they want to be contacted? How do they want to be contacted? That's that simple behavioral data. Now then, when the physician has a specific question about a patient, that might include clinical data, laboratory data as well.

Genomic parameters such as gene signatures, such as, um, the expression of certain receptors on a tumor cell, et cetera. It very much depends on the disease and the question the physician asks. 

[00:09:09] Kevin Folta: So has in silico discovery of different drugs, has it really changed a lot in terms of how drug discovery is done and what advantages does it have over what's been done Traditionally?

[00:09:23] Michael Zaiac: I, I think it has not revolutionized, um, drug discovery, but it has increased its efficiency. Greatly. So we use, of course, in silico, both in a, in a preclinical modeling of, of targets, how could our, or how likely our, our compounds hitting certain target. And then of course we are looking into expanding it more into the clinic.

And, and that goes back to my comment on, on culture and cultural change. I think the regulator has opened the door. I think US fda, um, has, or Congress has enshrined in law that, that US FDA has to look more. Into modeling system and so does the European regulator. And now that this store is open, we are looking more to try to also replace clinical studies with in silico trials.

However, of course, at the moment, that is very much an intent of the regulator and we as an industry have to come up with models. Which make the regulator certain enough to go for approval until that step happens. I believe in silico will really help us in the early setup of, of studies. It'll help us in the preclinical setting, but it will not in itself would place in human drug development at any point.

[00:11:10] Kevin Folta: Well, can we look a little bit more at the role of artificial intelligence in target identification and then really validation of drug development? Like how has, uh, artificial intelligence changed the, uh, efficiency and effectiveness of drug discovery? Yeah. For me 

[00:11:28] Michael Zaiac: it's, it's, I often, and, and this is probably reflecting my age, I often think of a microscope and if you think of a microscope, you have your ocular lens, I guess, which, which is usually a 10 times magnification, and then you have your revolving lenses.

I think normally times one, times 10 times. And now if I look at artificial intelligence, what it does help us is really to turn that revolving lens to gain much more insight into patients. So if we really start with that, we have the, the, the ability to gain insight from large, very, very imperfect dataset, which give us some patient characteristic.

Which might lead us to, to the better dosing of medicines, the better use of medicine, the identification of new targets in a certain disease. And if we turn the notch, say to enlargement 10. Then we might look at biomaterials of the patients. So these are laboratory data, these are cellular data. And here again, together with the disease outcome in these huge data sets, we can try to find, um, Novel targets, and if we go down to the modification that puts it all together and it puts together the bio signals we find together with the clinical outcomes and the clinical signal and the patient reported outcomes.

I think that in a, in an ideal world, will lead us towards personalized medicines. 

[00:13:21] Kevin Folta: And along the same line, how are real world data, real world evidence? How are they fitting into the scheme of regulatory and payer submission? And if, and if we look across the potential challenges and benefits of incorporating, uh, real world data, real world evidence, uh, what are the potential challenges?

[00:13:41] Michael Zaiac: I think real world evidence is for us, at least in Europe, um, probably more so than in the US today. Um, a standard add onto a payer submission packet. Whilst we have, um, emailed the European regulator, which is a little bit like the FDA covering a large geographic region of 27 countries. Um, Each of those countries has a different approach to the reimbursement of medicine, much like each HMO in the United States.

Um, and when we submit our packages, we often submit real world evidence. It might just be as simple as epidemiological data. On a certain disease, but it might be more complex, um, uh, illustrating the burden of disease, et cetera, et cetera. Um, regulators are perhaps where real world evidence is concerned, somewhat more conservative because their question is a question of, of efficacy, whereas, of course, the payer.

Ask for efficiency. Efficiency questions can be well answered with the real world data, uh, efficacy questions of a new medicine, lesser. So, um, so that's one of the limitation. The other limitation, and again, it's an experience we have very much in Europe and to a lesser degree in the United States, is, um, the availability of.

Quality data, um, that really limits how we can use real world data today. And part of that is due to the interpretation of privacy laws in selected European countries. 

[00:15:41] Kevin Folta: We're speaking with Dr. Michael Zak. He's the head of Medical Affairs Oncology in Novartis of Europe, and this is the Talking Biotech podcast by Col Collabora.

And we'll be back in just a moment.

And now we're back on collabs talking Biotech podcast. We're we're speaking with Dr. Michael Zak. He's the head of medical affairs oncology in Novartis of Europe. And we're talking about the modern changes that ia ai, it's artificial intelligence, it's, uh, artificial insemination and its active ingredients.

So I get confused. So, uh, we're talking about the changes in drug discovery and in the way it affects business in general with respect to ai and. How is AI being used nowadays to improve patient recruitment for clinical trials? Because this is always an issue. What impact has it had on the speed and success of trial enrollment?

[00:16:45] Michael Zaiac: Uh, I think we're just, uh, at the beginning and, um, let's see what we read from it in speed. Um, So what we, but, but not just we at Novartis, but many companies do, is of course we calculate the epidemiology of a certain disease for countries, for smaller regions of countries. Um, and from path site data, we then calculate the potential availability of patients, um, with a certain disease at a certain stage, um, at a certain site.

There might be some seasonality in that. Um, uh, and we look at that patient availability when compared to a broad spectrum of competitive trials running in this environment. So I think at the moment it really helps us to make some better informed decisions where to place our clinical studies. But the clinical trial business, at least in the area I'm working in, in.

It's very, very, Uh, competitive. So beyond having that knowledge, we still need to rely on the basics, which is to have an interesting medicine. You want to test an accessible protocol and appropriate remuneration for the investigators. Um, that still is the bigger. When compared to using AI and advanced technologies in selecting finding sites and recruiting studies quickly, 

[00:18:24] Kevin Folta: yeah.

But how do these digital technologies really give patients more choices, and how do they promote greater diversity in who participates in clinical? 

[00:18:34] Michael Zaiac: I, I think here, digital technology especially, um, driven by the step up change we saw in Covid, um, in the SARS CO two Abank. Uh, the pandemic, I guess no pandemic.

Um, we really saw a step up in decentralized clinical trial technology. Um, so we use a technology of, um, 15 modules which allow you basically to do the whole trial in the patient's. Um, now that sounds easy enough. And why haven't we done it before, and why would that bring more diversity? Well, patient Choice brings more diversity in the first place, but it also allows us to assess, especially also in the US C Communi.

Which are not necessarily going to hospital for a variety of good and bad reasons. Um, it allows us to assess rural communities. For instance, here in Europe, we pioneered it in Sweden, which is a country quite large geographically with a small population. Few centers accessible to rural population, uh, easily.

Um, and what we really saw is that we improved at least the rural city mix by using, uh, decentralized clinical studies. And really the hope is, and, and I think that's an observation again. The US FDA is that we will not just improve this rural city mix, but we will also improve, um, the ethnic mix, the genomics and US fda, I think in 2020 looked at their data and found that they had in the data submitted for.

Approval, um, they had 75% white Caucasian participants, of course, alone in the US. It should be no more than 60% white, Caucasian participants. If the trials were to reflect the US population. Um, so you can see that we still have a huge gap and we hope that decentralized technology is one step, but it'll only be one step to really move us forward in, um, better recruiting the right population.

Other things are, we really need to make sure that, um, The PIs, the investigators are of a diverse background, um, and we haven't placed too much attention into that, and we will do more of it. So that's, I think digitalization, digital technology, including blockchain, making it more secure really brings the opportunity of delivering trials in a patient.

When they want it, maybe one visit, they want to be at home, one visit, they want to come to the hospital. And that kind of choice attracts different patients, more diverse patients, um, opens it up to rural population, opens it up to population who have a natural aversion to, to, um, official institutions, et cetera.

[00:22:10] Kevin Folta: Well, can you really help me see inside the black box here a little bit and really describe the process of using AI to generate predictive biomarkers in personalized medicine. Like how does that work and how, how advanced is this field of oncology in developing targeted therapies? Great, Kevin, 

[00:22:29] Michael Zaiac: so I, I can give you a very simple example.

We started five years. Um, so we had a problem. We had a disease where you have too many red blood cells. Okay, sounds harmless enough, huh? Um, however, of course for some patients with these too many red blood cells, um, that means that there are arteries or veins can clock and they can have a stroke or a heart.

However, there's only a small amount of those prop uh, these patients, uh, every year say 3% maximum. How do we find those 3% of patients and give them more treatment than the other 90 plus percent of patients? That was our conundrum. There were some simple risk stratifications, such as H above and below 60.

Or if you had an arterial clocking before, that's not very good though. So we looked at a dataset of 90 million patients. We found 70,000 patients with that particular disease, and within a few months, AI software allowed us to find two simple laboratory parameters to identify those patients, more likely to get a clocking of the artery and more deservingly, if you wish, from a physician point, from a peer point, from a regulator point of view to receive high quality and probably also costly.

Or other interventions in this case, medicine. So I think this really brought personalized medicine to this particular disease, and it was only possible because we had artificial intelligence being able to crunch these large, very inconsistent data sets and making sense of sometimes. Less than superior quality of data.

[00:24:47] Kevin Folta: I guess my, uh, confusion on this comes from the fact that I don't know what kind of data are going into this. And you say you have AI crunch the numbers. What numbers are they looking at? 

[00:24:58] Michael Zaiac: They're looking at, um, every kind of patient data. In this case, we wanted to find simple parameters, so we didn't want to.

The genomic signature, we didn't have biomaterial. This is old patient data from 20, 30 years ago. Um, so what went into it is things like your blood pressure, your blood count, your lymphocyte count, the outcome of your disease, whether or not you had a stroke or a heart attack, all these kind. Quotation mark.

Very simple data went into the black box. 

[00:25:44] Kevin Folta: So, so we're really looking at very basic data and when we start to look at genomic overlays, not just, uh, DNA n a sequence, but gene expression, uh, you know, I mean, these kinds of things could be informing future ai, but it, I can see it's gonna go up exponentially and complex.

[00:26:03] Michael Zaiac: Correct. The complexity is going to be exponentially more. Um, and it may or may not bring us closer to some of the more complex question. This was a very simple question, or, well, for the patient, it's a very lifesaving question, but we wanted to resolve it very simply, and by chance we were lucky and we found two simple parameters.

Of course, we could have ended. Finding nothing in which case then you would resort to the next level, turning that revolving microscope towards genomic data, gene expression data, eventually proteomic data, which I believe will give you better insights even more. 

[00:26:52] Kevin Folta: With that said, what is the future of technology integration and pharmaceutical r and d look like?

Well, advancements in data collection and analytics really contribute to the development of truly personalized 

[00:27:04] Michael Zaiac: the. Absolutely. I think there is a nice saying being made about, um, a Harvard professor, about AI physicians and ai, and I think that seems true if you scale it up to pharmaceutical companies.

Pharmaceutical companies, um, will not be replaced by technology alone, but those companies not using this technology. Will become distinct in the past. I think it's, uh, you really have to be on board using digital technology, including advanced analytics or ai. I think one, one example is of reasoned in Europe is a, is a smaller biotech, which has been able to take a tumor sample from a patient's tumor.

Measure, its, um, antigenicity. So the ability of the tumor to trigger an immune response, predict the 10 most important antigens in that, and then construct the AI software then construct. As a vaccine against these 10 antigens. So each vaccine becomes personal to the patient with the tumor sample. And I think that's the first step in the right direction.

And of course, we will not see that happening for every disease because every disease of. Could be treated personally, like if you have a high blood pressure, hypertension, probably yours is different to mine and your, the treatment would be different. However, as the results with fairly non-personalized standard treatment are good, there's less unmet medical need where there is unmet medical need like an oncology, like an autoimmune diseases.

Wherever we have to modulate the immune system to either stop or um, uh, cause um, a reaction. I think there we will use these kind of personalization, including insights into genomics, into gene expression, into proteomic, um, into, um, post translational modifications, et cetera. 

[00:29:32] Kevin Folta: Well, how have all these digital technologies really impacted patient engagement and communication within medical affairs, especially say in the context of oncology?

[00:29:44] Michael Zaiac: I, I, I think digitalization and, um, the availability of information through omnichannels, um, have. Patient knowledge and patient education and therefore engagement, are we there yet that um, we can use them to communicate with the patient 100% better? No, not yet. I think there are many aspects we still have to.

Um, part of that is technology. Part of that is, again, the cultural change. What will society allow us, what will society see as just of pharmaceutical companies and engaging with patients? Um, but I think overall technology has opened the flood. To that engagement, both from the patient as well as former from a farmer industry.

Point of of you. 

[00:30:52] Kevin Folta: Well, that really does lead to the big question. So based on your experience at Novartis, if you were to stand back and give it that, you know, 10,000 foot view, what advice would you give to other companies in the biotech and pharmaceutical space that are looking to integrate digital technology into their r and d and medical affairs operations?

[00:31:13] Michael Zaiac: I think for me, the most important thing, Stay with your feet on the ground. Meaning, know what the question is. Don't develop technology in a free space. Um, meaning don't have centralized technology development. Don't, don't do things just because you can do them, do them to answer a customer question. So synchronize with your customer's.

And make sure that you meet those needs rather than to really free spirited, develop, develop, develop. Um, and the other thing is don't assume things will change because you have a great technology. Everything, even if you think it's great, does require a considerable effort in cultural change and implementation management to really be.

Otherwise, it's like the the best strategies they never implemented. Hence, we never know. Are they the best strategies 

[00:32:21] Kevin Folta: or not? No. Very good. So if people wanted to learn more about what's happening in ai, drug discovery in Novartis, where can listeners learn more? I mean either website, social. I think 

[00:32:34] Michael Zaiac: there's a Novartis website where you can see what, what we are doing.

But of course there are many other big pharmaceutical and and smaller pharmaceutical companies websites, and you can see what their tech, tech technologies are. I always found it extremely insightful to look. At, uh, the websites of the big consultancy firms like McKinsey, Pricewaterhouse, Coopers, and, and many others.

More interesting. I think there are also now really high quality executive education programs. Um, so if you want to go the whole nine yards, I guess, as you say, in the. It's, uh, there are education programs such as Howard and Wharton who look at disruptive digital technology, um, in corporate, uh, in the corporate world with a view to pharmaceuticals.

[00:33:28] Kevin Folta: Now do they say go the whole nine meters in Europe? 

[00:33:32] Michael Zaiac: Uh, we don't play American football so Well. I know my son is an avid football player. So. 

[00:33:39] Kevin Folta: Well, Dr. Michael Zak, thank you very much for your time today. I really appreciate it and well, we learned a little bit more about your company and, and how they adjust to changing time, so thank you very.

Thank you very much Kevin, and as always, thank you for listening to The Talking Biotech podcast. Write a review on iTunes, Stitcher, or wherever you consume podcast media, but most of all, share with a friend. Uh, we're coming up on 400 episodes going into, I think our ninth year or something like that. And, uh, It's really wonderful that you've been with us this long and lots more good stuff coming in the future.

So this is The Talking Biotech podcast by Collabora, and we'll talk to you again next week.