Quantum Shifts in Biotech: A Discussion with Zoran Krunic of Amgen

Interviewee: Zoran Krunic
Senior Manager of Data Science, Amgen
Interviewer: Tom Zuber
Managing Partner, Zuber Lawler

Details

Transcript

WE NOTE THAT THE FOLLOWING TRANSCRIPT WAS CREATED BY A ROBOT SO PLEASE FORGIVE ANY TYPOS.

[Tom Zuber]
Zoran. Hello, how are you? Good to see you.

[Zoran Krunic]
Very good. Thank you for inviting me.

[Tom Zuber]
Hello to those of you who are watching. My name is Tom Zuber. I’m the managing partner of Zuber Lawler, and I am on the editorial board of dead cat, live cat and online quantum computing news magazine. So Zoran, if you could tell us a bit about yourself and your role at Amgen, that would be a great place to start. Sure.

[Zoran Krunic]
So I joined Amgen six years ago as a senior manager of data science. The role was initially to introduce more of the machine learning, classical machine learning, into the clinical setting, patient data, clinical trial enrollment data, electronic health records. Also. Natural Language Processing was one of the very interesting new areas developing at the time, and so that became part of the work as well. And then shortly thereafter, we started working with quantum computing, which actually became my main work over the last four or five years. And so with generative AI coming in over the last couple of years on tail of the natural language processing, which was like natural progression of NLP into the generative AI, that happened probably about a year and a half ago, so that’s been roughly my journey with MJ. Oh,

[Tom Zuber]
That’s great. And I’ve been quite excited to speak with you, Zoran, as I think you know, to do this interview, because I personally have an interest in understanding how quantum technologies are being applied in the pharma space. I think the applications are, I think, obviously potentially dramatic, and we’d love to learn more about them. So let’s start with that. Can you tell us a bit about Amgen and its approach to this quantum conversation? And also, if possible, in as much detail as you feel comfortable disclosing any projects that you’re working on for Amgen involving quantum technologies, certainly.

[Zoran Krunic]
So. Amgen is a large multinational biopharmaceutical company headquartered here, just not from Los Angeles, in 1,000 Oaks. There are four general pillars that it has. It’s inflammation, disease, oncology, general medicine. And recently as of last year, rare diseases, after the large $27 billion acquisition of Horizon therapeutics, which brought rare disease molecules, molecule being really a medication, into the engine scope. And so as far as quantum if you look at the how biopharmaceutical process works, you have molecular part which starts the process of trying to find out new molecules. And it’s very interesting area. From the cost perspective and time duration, it’s usually somewhere about 25 to 30% although it depends on the disease, depends on the company to discover candidate molecules. And then you go into this biggest chunk, which is clinical trials, which is phase one, phase two clinical trials, it’s small number of patients, then progressing to phase three. And then it’s quite a complicated process, but complex. But in all of that, you have presence of data, often very small data sets. Oncology, rare diseases, bring a lot of small data sets. So they’re very interesting challenges where classical, quantum, classical machine learning has problems building good models. So we are hoping to see that QML quantum machine learning can do better in that context. And so that’s really important area, what we can do with the small data set? So let’s say phase one, phase two clinical trials can have 50 100 patients building classical machine learning model, and that is very difficult. And so that’s where we are looking at QML, and actually focused quite a bit over the last few years on that. So molecular part, we are mostly looking at small molecules right now. As far as quantum computing applications are concerned, there is a push to, of course, move to the bigger molecules. I think that territory is mostly done with llms large language models right now, since Google’s alphafold And, of course, there are all the classical approaches to molecular development that are present and but these are compliments. So I think we are going to see that quantum is going to go more into bigger molecules as the capacity increases, and that combination, also of quantum and ai llms, is also very interesting. It’s something that’s probably started last year, but we are really starting to see more of that work happening this year, and we can perhaps talk about it more in other questions, but I think that’s very interesting area. Also, when you go in the territory of clinical trials, you have optimizations. Clinical Trial enrollment, which is very challenging process, is a lot about optimizations. There is also Marko. In Monte Carlo, methods which are used for drug dose optimization also very interesting part of the drug design. And so last year, Nvidia published a paper on qmcmc, quantum, Markov chain, Monte Carlo, setting the stage for the development. It’s not really easy to apply it. And so we are looking at some of that as well. But those are interesting processes that all build into this picture of producing better medications and faster for patients. And so that’s really important,

[Tom Zuber]
that that’s very exciting, very helpful. Let’s get specific. Zoran, if we can, I’d like to touch on on the each of those areas, or at least some of those areas, one by one. Let’s start with research and development. How specifically now, if possible, if you’re comfortable doing so, how are is Amgen applying quantum technologies to the research and development process. So you talk about a molecule, and we’re applying quantum computing and AI to the study of the efficacy and the physiological effects of that model. What does that mean that you’re applying quantum technologies and AI to that research and development process, right?

[Zoran Krunic]
So we are not in production with quantum computing yet. So that’s important to understand that we are not there, and we are really in a stage where, after we started, we did quantum computing. About four or five years ago, there was a big contract with IBM. We did the collaboration. It was very good. We published the paper on that work. Now we are the position. I think we are not the only one is to try to move past the research and into the actual utility, where we can see that, maybe not today, but in 235, years, we can see the trajectory of improvement that eventually will cross the threshold. And we’re going to say, Okay, now, at that point, we can do things with quantum that we cannot do that with classical today, for example, going beyond simulations, which are usually about 2530 or 35 qubits, if the problem requires 50 or 60 qubits, that means you’re going to have to go on the real qpu platforms, which, of course, that implies some of the challenges there that are with data, with Error Corrections, noise, reproducibility and things like that, but I think the focus right now is finding the use cases that are more likely to have that positive trajectory over the next couple of years, and make sure that there is some kind of advantage, or at least that the quantum solution will become a useful complement to classical computing, because it doesn’t necessarily have to be better, but could be in some way, complementing and doing things that classical cannot do very well. Optimizations are interesting. I think also, let’s stop there,

[Tom Zuber]
if we can. Zoran so the quantum computing, the hope is we’ll be able to do things in the research and development space of the pharma industry better than classical computing. Can you explain that? Explain why that is why you think that is why you have that interspecific well,

[Zoran Krunic]
so if you look at the quantum computing and qubits in you look at the binary pattern. For example, in classical computing, all the bits are processed, all the patterns are processed sequentially, very fast, HPC with high performance computing, but still in quantum computing, all the bits and different possibilities are processed in parallel, because the numbers are not just zeros and ones, it’s actually probabilities that describe ratio zero and one in each qubit. And so when you look at 50 qubits that’s developed in parallel, that’s effectively, although it’s probably a simplified version, but we have computational capabilities in quantum computing that will, in effect be impossible to meet with even high performance computing today. Now, what we do see occasionally is that algorithms from quantum computing are becoming a motivation for classical computing to improve. So you see some of the advanced there as well, but it is really

[Tom Zuber]
If we could Zoran, and pardon me, but I really want to understand, because I really am personally curious the so we understand that when you’re when you’re dealing, obviously with qubits, as opposed to, I’ll call them typical bits, you’re able to Consider many different possibilities at once. Specifically, how are you anticipating applying that to the research and development process in in the in the pharma space? In other words, so what like, how is that going to I can extrapolate here, but I’d love to hear in your words, you know, where you’re not extrapolating. You’re doing or you’re planning on doing, how, are you going to use that capacity of quantum technology to improve the research and development process in regard to, let’s say, a particular molecule that you’re considering?

[Zoran Krunic]
Yeah, so I’m not a molecular expert, so I’ll probably skip on that. But even if you look at molecules the way the when you put atoms together and look at the quantum effects that really drive the molecular composition. I think that’s where there is a theory, and I think practice, that shows that the quantum effects are better described in quantum computing than with classical computing, although I think there are some very effective classical models as well, but one of the probably the simplest. Examples is if you look at optimizations. Optimizations underpin many types of calculations, even machine learning, even typical optimizations, when you optimize the enrollment of patients or logistics, for example as well. There are many, many problems that are way outside the possibilities of classical computing to NP, hard problems and a lot of examples that are already known. And so if you can solve those, let’s say, 50 or 100 qubits, and do it much faster and not I’m not talking about order of magnitude. I’m talking about number of orders of magnitude, then you have the ability to attack problems that are simply not possible today. So that computational capacity of of parallel execution of different patterns is, I think, the key very

[Tom Zuber]
good. And can you give us an example of a problem that that, that quantum computing, quantum technologies would would assist you with, assist your colleagues at Amgen with, that that traditional computers really don’t have the muscle to handle right?

[Zoran Krunic]
So, for example, if you look at clinical trial enrollment. That’s optimization problem. You design a new clinical trial and you want to deploy it in, let’s say, 50 countries, for example, across the world. And usually there are parameters of the design that says how many sites you have to go to. Site is clinic or hospital, and so becomes optimization problem, how you allocate your resources that when you open the site, that it’s most likely that you’re going to enroll the patients that you need, and cost wise and time wise, because time is really important component, you have a very complex optimization problem. Now, there are solutions today that do that and do that reasonably well, like genetic algorithms in some of those, but there are lots of hyper parameters of these models, which are not addressed today because of the limits of computational capacity, because you’re looking at optimizations that may take few hours or even a day, but if you add three or four different hyper parameters, each with couple of values, now you’re looking at the month, potentially or longer. And so to be able to expand into bigger hyper parameter space, that’s where the quantum computing will open up those possibilities, and that’s why you need those extra qubits to be able to do that and to paralyze that computing. So I think it’s that expansion of possibilities that are simply not done today in an optimal way. The other one, of course, is that in the optimism even your optimization theory, the landscape of the optimization space is often not known. It could be very rugged landscape with local minimums and global minimums, and you are trying to find that global minimum, I think quantum, and there is lots of theory on that, and lots of research paper that shows that those are areas where quantum should be able to do better. But again, classical computing has some solutions today, but I think ultimately the threshold, and there is a limit. There is a ceiling on that, that quantum computing, once you get qubits that are less noisy, that have less error, problems, that are more stable, you’ll be able to do that that classical just can’t handle. So it’s going to be a hybrid approach. I think that’s why even companies like IBM are looking at classical and quantum as a hybrid platform, where in real time, you decide where you go, classical or quantum, too.

[Tom Zuber]
I agree with you. I think that there’s going to be a soup of tools, if you will, and quantum is one of the going to be increasingly an important component of that soup. So that makes that makes good sense, and it’s very exciting. What’s the timeline Do you think to get to the point where Amgen is actually using quantum technology in this way.

[Zoran Krunic]
So I think that’s kind of a little bit of chicken and egg problem. I think lots of problems that are solved today, I are solved with the limited complexity, because that’s computation. What you can do, usually, when you have to start a process that’s going to run couple of days that starts to feel uncomfortable. It’s not practically very useful, right? So you’re always trying to stay in that range, or maybe few hours, maybe a day that your process finishes, because you have to run many times. And so I think there are lots of hyperparameters that will be added to processes, and quantum machine learning is the same way, although I think there are some nuances there that are slightly different. Is that by adding those and basically bringing more business value to the solution, making it more grounded in the complexities of the real world, I think you’re going to be able to do that. How many years? I think this is something that is matter of setting expectations. I don’t think anybody expects it to be this year, and I think we are hoping it’s not going to be 10. I think it’s probably a number in between. But every solution and every problem and every data, I think, will lead to different configurations and ultimately, different timelines. I don’t think there is a general solution to each of those to say, Okay, it’s going to be three years or five years. I think take

[Tom Zuber]
a guess, though, just because I’m going to I’d like to hear your best guess is to a year, three years, five years, some of three to

[Zoran Krunic]
five, somewhere in that range. I think there are solutions that we are looking extrapolation, the trajectory and, for example, imputations. There was very interesting paper that came out last year from QC ware. When using quantum computing for imputing data, which is a common problem, and everybody does imputations that based on the qubit projection, we probably need so many good quality qubits that we should be able to be there, let’s say, in two to three years from now. So I think that’s a guess of the trajectory, assuming that quantum physics and quantum hardware development really follows the trajectory so far. It seems to be doing that quite well. IBM is doing very well. Ionq, other companies as well. So indeed,

[Tom Zuber]
I think that that’s a reasonable guess. And I would have kicked out a similar, I think, range for what, what that’s worth. And I know we’re both guessing. Let’s talk about the impact. What kind of an impact do you think that this will have? Let’s let’s advance Fast forward five years. Let’s say you were correct. It happened in year four. What? What looks different in terms of the research and development process? How and how big is that difference?

[Zoran Krunic]
Well, I think so everybody’s hoping that you may have a chat GPT moment, but I think it’s better if you’re doing incrementally, because you have more control over the process. And so I think bringing more complex cases under the computing paradigm that could be executed and simulated will allow us to develop better enrollment clinical trials, for example, better molecules. And so you’re shortening the process the that from the moment you start to the moment medications get approved and reach the patient. And that’s very important. Obviously, that’s that’s very easy to understand, the ability to generate better molecules that have less attrition in the process of clinical trials, because lots of molecules, when you start, usually only a couple of percent make it to the end to actually become medications that are approved. Lots of them fall out. And so I think that’s the part where we expect that to be more efficient means cheaper, and that, of course, that has downstream positive consequences of the on the on the medications themselves, on patients, and everything else, the research itself, the cost of research. So all of that comes together. Then

[Tom Zuber]
Zune, what are the biggest obstacles in your way? In Amgen’s way, in the in the pharmaceutical industry’s way to to making it to that auspicious day?

[Zoran Krunic]
So I think we are very rarely in quantum I think there is very little knowledge about quantum computing, what it does, what it can do, expectations. It’s not that widely available. There is a cost factor as well. The QP time is quite expensive. And so I think those are factors that right now the one of the biggest challenges is, and I see two components there. One is strategy, let’s say the five year strategy. How you do that building, I think, for the moment, centralized quantum computing, teams that work on that in completely dedicated fashion. Somebody once said, about four years ago, five years ago, you started QC. Quantum Computing is not a hobby. You don’t do that on Friday afternoon only, right? It is something that you do full time, and you need more demo press. You need centralized quantum computing teams for now, later, you may decentralize part of that, perhaps because you need to be connected with business. That’s a really important connection. So some hybrid but I think that’s the one education of enterprise, human structure, I think it’s important so that lots of people, employees, managers, directors and above, understand actually what’s reasonable to expect, what’s possible, what quantum computing, computing is all about, and find those better use cases which everybody is really looking for right now. So I think those are the two budget, of course, is always a factor. You need to allocate some budget, because that’s going to be important. So I think those are the three main ones, education strategy and resources on a full time basis with some financial support. Makes

[Tom Zuber]
good sense. I will point out that you mentioned that quantum computing isn’t something you pursue as a hobby as I pursue my hobby of interviewing folks on quantum technology. So on that note of just, you know what we do for a living, and tell us, how does a data guy, I can guess, but how does a data guy get into quantum computing at Amgen, right?

[Zoran Krunic]
So I my, you know, I come from computer science background and masters in quantum computing. Also, I was in data background for quite a long time. But Analytics as well, I’m interested in their other areas, different technologies like evolvement, even chaos theory and things like that. So they all play a role, I think, in building the bigger picture. But ultimately, if you look at analytics for last, let’s say, 30 years, with corporate data warehouses and enterprise data warehouses becoming dominant data platform for for enterprise environments, and probably the largest investments ever made, not just one time, but as a continuous investment of bringing business processes through the data into stable and high quality repositories, that’s the. Probably the most important component of all the analytics, if you don’t do that well the downstream Analytics, you can bring quantum you can bring llms and AI, nothing is going to help you. So I think the strong quality of that is something that resonates with me. I think classical machine learning is very important as well, but those are the two. And so when quantum computing came, I looked at that as a new type of analytics, but still anchored in the same process, in the same data. And I think from that perspective, I was curious to see computation, what it can do, and became actually very interesting area that I pretty much spent most of my time over the last five years.

[Tom Zuber]
Zoran, that was fabulous. Thank you very much. And for me, it’s a great gift to be able to speak with somebody like you. I like to interview folks, but I I don’t have the capacity or the time to do what you do, and I think it’s very exciting. And I appreciate the education very much. To be able to get a small peek at the future through your eyes is really a gift. So thank you very much, Zoran for taking the time. Greatly appreciate it.

[Zoran Krunic]
Thank you. Thank you for inviting me.

 

Zoran Krunic - Guest

Senior Manager of Data Science, Amgen

Since joining Amgen R&D in 2018, Zoran Krunic has been at the forefront of applying Machine Learning to enhance patient outcomes and streamline clinical trial enrollment processes, utilizing comprehensive Electronic Health Records and clinical datasets. His pioneering work in the Quantum Machine Learning space, in collaboration with IBM’s Quantum team, has been instrumental in integrating machine learning with quantum computing through IBM’s Qiskit platform. Prior to his tenure at Amgen, Zoran developed Machine Learning algorithms at Optum to predict hardware and software failures within complex enterprise architectures. He has a strong background in data engineering and systems development, having contributed significantly to large-scale projects at renowned organizations such as Capital Group and ARCO Petroleum. In his current full and part-time endeavors, Zoran is leading the efforts in embracing generative AI technologies, with a particular focus on OpenAI’s GPT and Anthropic’s Claude models. His work is focused on prompt engineering and its application to code generation, advanced document analysis, and process management, with a commitment to ethical AI practices and data privacy. Most recently, the work has expanded to include reproducibility and the potential for auto-labeling using generative AI. A recognized voice in quantum computing circles, Zoran is a regular presenter at industry conferences and has served on numerous panels discussing the integration of quantum computing and generative AI within the Health Sciences sector, with a firm belief that the integration of these new technologies will be deep and comprehensive. With a Master of Science in Electrical Engineering & Computer Science, Zoran continues to explore and contribute to the evolving relationship between quantum computing and artificial intelligence, fostering groundbreaking advancements in healthcare technology.

Tom Zuber - Host

Managing Partner, Zuber Lawler

Tom is recognized nationwide by Chambers USA. He focuses on intellectual property litigation and deals, and has advised some of the most iconic companies in the world on their global intellectual property portfolios. He is the relationship partner for some 10 Fortune 1000 clients, as well as iconic government entities. He is the Managing Partner of Zuber Lawler, which advises clients around the world from offices across the U.S. as to M&A, finance, real estate, corporate, employment, bankruptcy, commercial litigation, and internal investigation matters. Tom is an avid futurist immersed in emerging industries and technologies, including quantum computing, blockchain, and AI. He holds a J.D. from Columbia Law School, an M.P.P. from the Harvard School of Government, and a B.S. summa cum laude in biomedical engineering from Rutgers University.

Leave a Reply

Your email address will not be published. Required fields are marked *

4 + 16 =