Decoding Cancer: AI, Biomarkers & Precision Medicine

[00:00:00] Dear listeners, welcome to Faces of Digital Health, a podcast about digital health and how healthcare systems around the world adopt technology with me, Tjasa Zajc. When it comes to cancer, prevalence is increasing and there's still a lot we don't understand about the factors and causes of cancer. Many studies have clearly demonstrated the benefit of biomarker testing for cancer therapy.

[00:00:30] However, broadly speaking, roughly 30% of cancer patients are eligible for targeted therapies based on their tumor profile. And even when the biomarker is present, roughly 30% of the eligible patients respond to these treatments. We have a lot more to uncover.

[00:00:52] So, in the discussion that you are about to hear, we're going to focus on the progress of precision medicine in oncology and the impact of AI in the endeavor of making cancer treatments more specific. I spoke with Luka Ausets, an expert in the field of biology and computational science, who works as the Chief Discovery Officer at Genialis,

[00:01:20] an RNA biomarker company which develops and validates clinically actionable biomarkers informed by the world's most ethnographically diverse cancer data sets to better predict patient responses and guide treatment decisions for targeted inhibitors, immunotherapies, and other emerging therapeutic classes.

[00:01:42] Luka oversees the internal research and development and external partner projects with the common goal of advancing therapeutic discovery through the rigorous application of data science. He believes a successful discovery process is built on clear lines of communication and unwavering scientific integrity.

[00:02:03] We discussed the current state of cancer research, role of computational science in drug discovery, clinical decision-supported development, and response predictions development in the field of cancer. Enjoy the show. Enjoy the show. And if you haven't yet, make sure to check out our newsletter. You can find it at FODH.substack.com. That's FODH.substack.com.

[00:02:30] I'll make sure to link to an edition where I already covered cancer research in the past. And if you have a minute or a few seconds, do leave a rating or a review wherever you listen to your podcast. It gives us the feedback on the work we do. You can leave us suggestions on the topics that you would still like to explore. And this will also help others find the show as well. So thank you.

[00:02:59] Now let's dive in. So Luka, thank you so much for taking the time for a brief discussion on how technology is trying to help us understand cancer and develop new cancer treatments.

[00:03:27] You're a chief discovery officer at Genialis. So let's start with that. What does that mean? What does it mean to be a chief discovery officer? Yeah, thanks, Tjaša. Thanks for having me. It's a pleasure. Chief discovery officer, in my case, oversees the biomedical aspects of our development. I also participate in disseminating our innovations and also protecting them. That's a major part of my work these days. You're a biologist by background, right?

[00:03:57] Yeah, that's correct. So what kind of fascinates you most when it comes to cancers and how our understanding of cancer is changing? Yeah, I think a lot of different important discoveries came about in the past several years.

[00:04:15] One thing that's really all over the place if you go to any of the more academic cancer conferences would probably be methods that try to dissect the tumors in 3D. So the spatial resolution of cancer tumors. We now know that the tumor is not just cancer cells, obviously, right? But the tumor microenvironment is very diverse.

[00:04:43] There are blood cells, immune cells, connective tissue cells. And we are finding now that these have much deeper implications in cancer development than previously thought. And Nature, the famous journal, announced spatial transcriptomics as the method of the year. I think it was 2020. And last year, spatial proteomics was named the method of the year.

[00:05:11] So it's really this 3D resolution when you can slice the tumor really finely and then observe how genes are expressed and how proteins are expressed in cellular and sometimes subcellular levels. So it's quite fascinating. Very sophisticated, very detailed.

[00:05:32] And that's why a lot of just the cancer research and trying to figure out which treatment is going to fit to which patient is the key in the current development of the field. So can you share a little bit on just what Genialis does in this sense of which aspects do you cover and focus on and we'll go from there? Sure. Sure.

[00:06:01] So for Genialis, we are a precision medicine company. And our goal is really to find or marry the right patient to the right drug and have precision medicine improve lives of individuals, but also their families and communities. So we do this by studying gene expression.

[00:06:27] So not only the DNA level, so the blueprint of our cells, but also what is actually expressed at any given moment. And by studying that at the RNA level, we can get a much deeper insight into the activities in the cells. And thus, hopefully, we can better predict whether a certain patient will respond to the treatment or not and also why. Right?

[00:06:56] So it's not just a black box, but really trying to uncover the mechanistic reasons behind how people react to drugs and how their diseases unfold. It sounds ideal that just by getting the genetic footprint of an individual, you could get all the information about which treatment is going to work for you specifically. But can you maybe elaborate a little bit on how that works in practice?

[00:07:23] And what I mean by that is we, at the moment, still know a relatively small amount about genetics. There's still so much that we don't know. So the fact that we can do whole genome sequencing is just the beginning of the story. So knowing that basically we know a little bit about the genomics and genetics of an individual,

[00:07:49] how, like, how do you, to what do you actually match that footprint? Like, how much research is that? How much research do we see is still happening to just find those correlations? Because when you're a patient, we are not in a research process necessarily where your data would be incorporated into a larger pool of existing patients or past patients.

[00:08:14] Yeah, this is all rather complex because what we're dealing with in our particular case is the small data problem. So everybody's about the big data, right? And that those certainly exist in the medical sphere. When you take electronic medical records, when you take your blood readouts, things like that are done en masse. Those certainly exist in large amounts, but we are working with molecular data.

[00:08:44] So DNA and RNA sequencing data. And those really, those types of data really aren't so abundant. So what we typically have is data of the wrong shape. We are collecting gene expression values, maybe 60,000 genes per patient, but we only have 100 patients, right? So the problem then becomes, how can we learn anything from this kind of data without overfitting to just those couple of responders?

[00:09:14] And the way we go about that is by applying biology to reduce this feature space. In other words, when we say, ah, we understand how this particular drug is supposed to work. We understand a little bit about the current biological underpinnings of this particular tumor. Can we put those together and still have a meaningful prediction?

[00:09:37] One that's robust enough that will be relevant for the next clinical trial or for the next patients when this comes into regular use. And frankly, we have been quite successful with that. The success is never cease to surprise me because exactly the reasons you just enumerated. It's quite impressive that something like that can be done, but it can only be done if we're taking into account bigger numbers of genes.

[00:10:05] So we're talking dozens or sometimes hundreds of genes and the intricate nonlinear connections between those genes, right? So it's not just about mutation, yes or no. It's a lot more complex than that. Mm-hmm. So how does that tie into treatments? Because in oncology, there's a whole specter of treatments.

[00:10:34] Surgery, still the number one. Then there's chemotherapy that's developing, radiotherapy, radioligand therapy, immunotherapy, probably I forgot a few more. And increasingly, when clinicians decide how the patient is going to be treated, they choose a combination of all these treatments.

[00:10:56] So how does that impact you and the choice of all the choices for treatment that the clinicians have? The way I think about the medical science, right? This will be maybe unorthodox, but it's less of a science than many people think, right? It's a collection of empirical knowledge, right?

[00:11:22] On average, if we give this person this type of chemo, they will be better, right? But that only works on average, right? And even that, not all that well, right? Ideally, that's the goal of precision medicine is to have some data-driven evidence or decision-making system that would say,

[00:11:44] this person will respond to this type of chemo or not, rather than just this intangible experience of an experienced doctor, say. So, ideally, and that's something that our company aims to do is figure out biomarkers for all these kinds of treatments, right? Even your standard chemo, right? It's not every chemo is for every patient, right?

[00:12:11] We know that already, but we would like to quantify that and would like to create models that can predict which patients would actually benefit from each of these very established treatments. And when it comes to more innovative, novel treatments, right?

[00:12:30] What biomarker-driven strategy can help is really figure out the right patient population for those drugs so that we can make the clinical development viable, right? Maybe your drug is great, but if it only helps 17% of the people, then likely you won't get the funds to actually develop the drug.

[00:12:53] But if we can enrich the patient population to say 60%, 70%, 80% responders, then this becomes a viable drug. So, it's not only that biomarkers limit the market for the drug. It's actually, do you want this drug to succeed or not, right? And if yes, these are the patients that you need.

[00:13:14] So, I'm sure that in time we'll have much more complex and much more robust biomarkers for many of these treatments. And markers that will not just be a single mutation, but will be really about the biology for that patient at that moment. How far is the field of biomarkers at the moment? Would you say, like, how well developed it is?

[00:13:42] How many biomarkers do we already know exist and are used? Yeah. I think that when you say biomarkers, a lot of people will have a rather narrow understanding of the field, right? So, maybe there's a certain target that's expressed or not expressed, or there's a certain gene that's damaged or not damaged. Simple things like that.

[00:14:06] So, I think we've got quite a good handle on those kind of very simplistic markers that are sometimes predictive of response to treatment, sometimes prognostic of survival, things like that. But to understand the complexity of the biology better, we'll have to come up with more complex markers.

[00:14:29] Ones that take dozens or hundreds of genes into account, their complex interactions, and so forth. And quite frankly, there's really a long way to go there. But ideally, it goes beyond that still. So, it's the social factors, it's the other determinants such as environmental food intake, things like that,

[00:14:54] that all have something to do with both occurrence of cancer as well as treatment of cancer. And we have not yet even started as a science, as a humanity at all, to really grasp with the complexities of that. Absolutely. I think in the movie documentary, Longevity Hackers, they mentioned that basically genetics present just 7% of the whole impact on longevity.

[00:15:23] Everything else is other factors that you potentially can impact. Yeah, we have a lot more to improve in understanding. We just... Yeah, here I'll just add, I think this is one of the very common misconceptions that, you know, that person is a really active sports person and still they got cancer or my great uncle smoked for 60 years and didn't have cancer and so on.

[00:15:49] I think this kind of examples undermine our appreciation of the extent that the environmental and societal factors have on cancer development. I think this is very much overlooked. And we have beautiful examples now of... Her name is Vargo from MD Anderson, like a famous oncologist. And she treats patients with very standard oncological procedures, but she splits them in two groups.

[00:16:18] And the difference between the two is the diet. She just feeds one group a ton of dietary fiber and she gets tremendous results. She's on every major cancer conference always because people are never tired of being amazed at the successes that she has in the field just for sticking chia seeds into their every meal, patients every meal. Right?

[00:16:44] It's a caricature obviously, but like the things like that have an enormous impact. Is there a living plant on your bedside table or not? Things like that, I think we're just finding out how important those things are. How do you observe the whole field of research and trying to understand all those factors better, especially in the light of the current changes that we see in the US?

[00:17:12] The government or the current president wanted to slash the research funding by half. Now that has been put on pause at the moment, so we don't know yet what is going to happen with the funding. But the way that we've described the complexity, it seems that we need to just research a lot and we will not know what the actual return on investment for that research is going to be.

[00:17:40] It's similar with drug development where only a very small percentage of the drug actually gets to the last stage of clinical trials and on the market. Yeah. How do you just see the whole field? I assume it also impacts Genialis as a company in terms of who uses your solutions? Yeah, it's a broad topic and one that's difficult for me to discuss calmly, right?

[00:18:09] Certainly a lot of uncertainty right now with the new president and the new administration coming in. I just attended a webinar where they discussed the recent drafts that the FDA put out for management of AI components in either devices or drug treatments. And it's good documents. They've been in the works for many years, right?

[00:18:36] Right now, their destiny is really uncertain. What is that going to... Are they going to reject everything? Do we start from scratch? The definition of AI has been rejected, the one from two years ago. So what are the grounds that we have to even discuss this kind of subject?

[00:18:56] So right now we're obviously paying very close attention to this kind of maneuvers and it's really just a holding pattern. You can't do much other than sit back and see how the chaos unravels, right? But it's certainly a confusing situation. Yeah, absolutely.

[00:19:21] So let's try to explain a little bit more the role of AI in the understanding of cancer and predictions that you are also helping to make. Which basically type of AI machine learning and other types do you use? And how did the whole emergence of generative AI look like? So how does it help you in developing your models?

[00:19:48] Yeah, as a user of generative AI, as a scientist, it helps me digest more of scientific literature faster. But other than that, I haven't really noticed much impact of gen AI on our work. What we do is predictive AI and arguably this is still the part of AI where a lot of impact and money is made, right?

[00:20:16] And I think it's really important to integrate in finance, in insurance, in medicine as well. So trying to have the machine learn from the patterns from large amounts of data and help you make some predictions. This is the business we're in. So I'm absolutely fascinated, obviously, by Gen AI. And I think it's really awesome to observe that.

[00:20:44] But there's an observation that many, also my colleagues in the office share, that humans are turning into interfaces that sit between two chatbots discussing things, right? I might create a project plan, right? And it has to be 20 pages and have all these different sections. And so what I do is I type in some notes, some bullets and have chat GPT or whatever other AI bot develop the text for me.

[00:21:13] And then the reviewer obviously won't read the whole 20 pages. They'll just stick that into a chat bot and have that summarized into a few bullet points, right? And he'll craft the response in the same way. So is it really me and the reviewer talking or is the two chatbots talking? There are certainly areas where I'm skeptical of the promises that this is going to bring in medicine in particular. Obviously, we saw a lot of those advances in other fields.

[00:21:43] I think for quite some time we'll still be, still see that predictive AI is where it's at. Yeah. Interesting comment. I was just thinking about this when I was driving here and I was basically talking to chat GPT to just tell me a little bit more about the history of chemotherapy and to name all the treatments that exist for cancer. And the models are improving. The hallucination rates are also decreasing.

[00:22:12] But still, the amount of data that AI is able and these bots are able to condense is so large that it's impossible for us to really check the accuracy of what we get, but we take it for granted.

[00:22:27] In your case, when you were talking about that back and forth communication with constant summarizations and interpretations, it almost seems as if in the worst case scenario, we could set back the science because we would be skipping between wrong interpretations or misinterpretations. Yeah, I think that's certainly possible.

[00:22:51] I would say that in science there's always going to be curious people that will want to learn everything and so on. So basically, I can now use a chat bot to help me summarize a paper just to help me understand better its contents. But if I find it interesting, I'm going to read it and I'm going to read the supplementaries and I'm going to read everything, right? Because that's still how the science gets done.

[00:23:18] So I'm not really worried that a lot of that will be missed, right? So it just helps me to sift through a lot of documents faster. As for AI predicting and replacing doctors and so on, I don't see that at all happening. Maybe not at all, not even anytime soon, but maybe not at all. Yeah, absolutely.

[00:23:42] And maybe we have a best case scenario and the predicted shortage of 10 million doctors that we're supposed to have by 2030 is not going to exist because we're going to able to optimize processes. For anyone working in healthcare, we know that's pretty wishful thinking because there's so many things that we still need to do in healthcare optimization.

[00:24:07] But if we go back to cancer a little bit, can we specify, for example, where do computational models, where have they been shown to have most success? So are there any specific cancers, any specific treatments that you can mention that have been, that profited most from this at the moment? Yeah, I could probably name quite a few.

[00:24:35] Certainly there are areas where we have very good targeted treatments, but those treatments only maybe help a minority of population, right? Finding those patients that actually would respond, I think that is a classic use case. That's both for drugs that are already on the market as well as for the newcomers, right? If you want to differentiate in a crowded market where similar types of drugs already exist, if you can pinpoint your population better, that's great, right?

[00:25:05] Which cancers, the major ones, gastrointestinal tract, lungs, breast, are certainly the cancers that have received more attention. And then you have more aggressive drug cancers, PDA, pancreatic, where not a lot of data exists, right? So this or certain urotelial or something like that.

[00:25:27] So in these cases, working with specific centers dedicated to those cancers will be paramount to just ensure that enough data even gets collected. For example, we have a very good relationship with a patient advocacy group, PANCAN, that does even more than that.

[00:25:47] And they also collect their own data. And we have quite a fruitful collaboration going in pancreatic cancer because, frankly, this type of data is quite difficult to get by, right? I think so that's one aspect, like which cancer types. But the other aspect I think that we're always talking about it, but we aren't doing much about it, is the demographic diversity, right? Are we really collecting all those populations that we need to be collecting, right?

[00:26:16] Lung cancer in China is quite different from lung cancer in Europe or in the US, right? That's because of the different genetic background of the populations and so forth. Are we really sampling all those different demographics, right? This is something that is very important to us at Genialis. You alluded to the broader issue of small data, which you mentioned in the beginning.

[00:26:43] What do you hope to see and what's the role that you see in synthetic data and digital twins to address this issue? I think this is a marvelous idea and really underused at the moment, right?

[00:27:02] Because in Slovenia, for example, the legislation governing personal data is much stricter than the general European regulatory environment, right? And it's like that in many places in the world. And this makes it really difficult to get hold of actual clinical data. But that data can be modeled, right?

[00:27:26] And I can just get the modeled synthetical data out that models the properties of that data, but I can never link it back to individuals. This idea would just free up so much data and have it available for researchers and companies to use to advance this.

[00:27:44] I think this is severely underutilized at this moment and we're bogged down by all the regulatory limitations of accessing this data.

[00:28:01] This really hampers research. I think if we can do something about that and synthetic data certainly is one of the possibilities, then this will radically speed up the AI incorporation into research and diagnostics or whatever. Really, data is crucial and it's so hard to get hold of data, right? You get a hundred patients, you're over the moon, right? It's awesome, right? A hundred patients is nothing. I need 10,000.

[00:28:29] What's the plan for me to get to 10,000 patients? So how do you get to patients and patient data? In your case, as a vendor working on the market, we're seeing in Europe the development of the European health data space. And the hope is that we're going to have this huge research pool. But then again, there's going to be a lot of limitations who can actually do research and what type of research. So how do you solve that issue? But at least there will be a common infrastructure.

[00:28:57] There'll be a common, a singular body that makes those decisions, right? Like an ethical committee for data. That's what I imagine at this moment, right? But it's coming in what, six years. So it's a bit of a wait, right? It's going to be a minute. So right now, what is left is there are public repositories of data.

[00:29:20] There are academic centers, which is something we pay quite a lot of attention to in forging those relationships with interested doctors and institutions that are trying to find ways within the current regulatory limitations, right? To share the data and collaborate on the data.

[00:29:43] And we do that in a lot of places in Asia, for example, the Middle East, India, Taiwan. There are patient advocacy groups that mentioned, Pencan, for example. And there are big data brokers, companies like Tempus that do a lot of sequencing for all kinds of reasons. And then they can do different things with those data, right?

[00:30:09] So it's all these things combined that get you to a certain, over the threshold that you can start doing something with the data. And then it's about connecting with the pharma that are running trials in a particular disease, in a particular stage, with a particular mechanism and so on, and try to work with them to get those data and so forth, right?

[00:30:35] It's absolutely number one limiting factor for us, just access to the data. How does the findings that you come across, how does that trickle down to the patient? For example, does the pharma industry use the models in their research? Where do they find something that's meaningful for the patients?

[00:30:58] And they add that information into the drug information so that clinicians can then see what they should be mindful of or what they should test patients for in order to see if that patient is going to respond to a drug or not.

[00:31:13] So let's talk a bit about that, because on the one hand, we're discussing how researchers can use just clinical decision support systems to discover new things or clinicians to decide what they're going to give to patients. And I just want to understand how that impacts the patient after all. Yeah.

[00:31:38] For models that we're creating, model certain parts of cancer biology that, and those models help us, for example, predict response to treatment. They also tell us why a certain patient responded and why not. Sometimes they give us new hypothesis. Sometimes they give us outright explanations. But it's often more than just a simple prediction, right? If A, then B or something like that. It's often more complex.

[00:32:07] So it's really about the mechanistic insights. When we say biomarker, sometimes people get bogged down with the narrow definition in their mind, but it's really about know your drug, know your patient kind of scenario. Whereas, where we want to, so right now we're working mostly with pharma and sometimes in very early research stages as well. Right?

[00:32:36] So we're not necessarily talking about, oh, this will be a companion diagnostics device that will be used with the drugs. So whenever, you know, a patient will be tested with this particular test and it will tell them yes or no for the drug. We're not necessarily talking about diagnostic devices or those are certainly important. But even earlier on, it's important for researchers and pharma to understand what are really the eligible populations for a particular drug.

[00:33:05] What are adverse events? Why they happen? In what populations? And so forth. So this improves the safety profile of the drug as a whole, even if in the end there isn't a diagnostic device associated with the drug. So this is an indirect answer. So how to your question, how does that trickle down to the patient?

[00:33:27] First and foremost, that so that the drugs that come to the market are better drugs because their scope is better defined. What do you look forward to most when it comes to cancer research and figuring out how can we improve? Yeah, basically cancer treatments because the prevalence of cancer, if you look at predictions, isn't really something one would want to know.

[00:33:55] I know that for Slovenia, one in three women and one in two men born in 2019 can expect to get cancer by the time they're 75, which sounds awful. It also sounds a bit less awful if you say 30% of women or 50% of men versus one out of two. But yeah, so we are definitely not nearing the situation where we would be able to cure cancer.

[00:34:24] So what do you, what keeps you optimistic around the field? So sometimes I get a little bit disillusioned by the kinds of data that you just mentioned, because sometimes it just seems that we're spending a whole lot of money to extend old people's lives by three months, right? Or something like that. But young people get cancer, right?

[00:34:55] When I see in our spreadsheet that a 40 year old mother returned to their family after a successful treatment with a novel drug, like that is really meaningful to me. So I think understanding patients and figuring the, really the promise of precision medicine is a fantastic one, right? Really figuring out how to marry the right drug with the right patient.

[00:35:21] That said, I'm, I don't think we're going to cure cancer, right? I don't think such a thing exists, right? First of all, cancer is so many different diseases. Second of all is yes, some things can't be cured. It's just like that. If you're, it's harsh truth. And earlier I told you about the mission of Genialis, which is one where precision medicine rules and delivers best possible outcomes for patient families and communities.

[00:35:51] I think the families and communities bit is one that's super important. So while the biological, biomedical science and research can figure out the individual bit better, right? So which really is the drug that you should take at this very moment? The social part of the equation, right?

[00:36:14] The families and the communities bit is something that it's not really for us biologists to figure out, right? Is it, you'll know these numbers better than I, but is it, is it okay for us to spend 85% of, of medical budget on a person that's the whole life, right? Budget for the last year of their lives, right?

[00:36:39] Is that ethical? I don't know. But my, my, my feeling is that a lot of this social engineering will be needed in order to make precision medicine realistic. Not everybody gets the best treatment, right? The best treatment for the individual might not be the best treatment for the families and the communities, right?

[00:37:05] If this is very difficult to, to talk about in, in just a very short amount of time, but I think this really is realistic, right? Saving lives at all costs is, is just not the solution. Yeah, it's a broad societal topic, especially in the era where sustainability and costs are a big issue for healthcare. So we should, we could have a different, another whole discussion around that.

[00:37:34] Yeah. I'm certainly not an expert in the area and there's a mire of ethical considerations there that I'm aware and maybe not aware of. But I think these are also the kinds of discussions that we, we need to have, right? It's not just about advancing molecular biology and, and dissecting it further and being able to put on a VR goggles and 3d explore the tumor and CD, whatever, right?

[00:38:05] And then, and then suddenly, and then suddenly, and everything suddenly magically unravels. I don't see that happening. And I don't think that's the best also for the families and communities bit of the equation. Yeah. Yeah. Yeah. This reminds me of a discussion that I had with a nephrologist from Nigeria, where resources are obviously much scarcer than here.

[00:38:31] And I remember what a cultural shock it was to me when she said that because they've got limited dialysis machines, they first of all, give less, give dialysis less often than we would in the West. And secondly, sometimes they would decide to put a patient directly into palliative care because there's a younger patient that has a higher chance of survival.

[00:38:58] So these are the types of tough choices that are definitely not necessarily discussed when we talk about digital health and technology and its promises. But yeah, I hope that we see more of that optimism, as you mentioned before, of those stories where you can make a huge difference on somebody that maybe is younger. Is there anything else that you would like to add in terms of what has been happening to you as a scientist, as a researcher in the field?

[00:39:28] Anything that inspires you or any last thought that you might have? I think it's certainly a very exciting time to be alive, right? For a researcher in our field because of all these advances in AI, just the amounts of data, the methods that we can use to analyze those data. So it is really exciting.

[00:39:56] I can mention things from AlphaFold to cell-free DNA to all these kinds of things that have suddenly appeared, obviously not out of nowhere, but... Suddenly have found real use in both research and application of medicine. I'm looking forward to see some more of that in the future.

[00:40:22] You've been listening to Faces of Digital Health, a proud member of the Health Podcast Network. If you enjoyed the show, do leave a rating or a review wherever you get your podcasts, subscribe to the show, or follow us on LinkedIn. Additionally, check out our newsletter. You can find it at FODH.substack.com. That's FODH.substack.com. Stay tuned!