Latest / Diagnosing Dx / Can AI Rescue Cancer Drugs That Clinical Trials Say Have Failed?
Transcript
- David Filby: Sasha, welcome to the show this week. How are you? Alexander Bagaev: I'm great. Thank you so much for inviting me. David Filby: Absolutely. Where in the world are you? Alexander Bagaev: I'm in Boston right now, enjoying the heat wave. David Filby: Good for you. And the World Cup. Alexander Bagaev: I'm not a great soccer fan, so it's all my friends cheering and inviting me over, but I'm skipping. ⁓ David Filby: There we go. Well we've got some awesome stuff to get into today. I'm really excited about getting to know more about BostonGene, but before we do that it'd be great to learn about yourself, what your background and career is and an introduction to yourself. Alexander Bagaev: Thank you so much. So my name is Alexander Bagaev or everyone calls me Sasha. That's a usual interesting change my name, but for Russians that's kind of obvious. That's a very popular ⁓ Russian nickname for Alexander. I did my master's in physics and Initially I got a pretty deep technical education and then I switched to the biology. I did my PhD in immunology in Moscow Institute of Immunology. but because I was planning to be a physicist, but ⁓ I realized that's ⁓ that's too complicated in biology. It looks far more interesting to too many unknowns. The physics helped me understand how to think in terms of modeling, how to apply mathematics to biology, and had that naive thought that we can easily apply math to the biology. ⁓ I was very naive. after that, all my career I've been focusing on understanding the complex patient's biology through the prism of genomics, transcriptomics, and imaging. ⁓ so I've been focusing on developing algorithms to integrate different data together to understand the holistic view of the patient tumor and immune So I joined BostonGene corporation. where focused primarily on oncology and oncoimmunology. So there I started my path towards developing the first genetic product to analyze the RNA DNA sequencing data to help patients ⁓ select right treatment. Along the way building software and basic scientist teams, machine learning teams, ⁓ it's AI, to analyze and understand complex and make it clinically useful. So that's that's a short story of myself. David Filby: How long have you been with BostonGene now? Alexander Bagaev: I've been about 11 David Filby: And initially in Moscow and then over to the US, is that right? Alexander Bagaev: That's right. That's right. I was hired for the Moscow office and I was building the Moscow office by Moscow, good math, good software back the days, good access to greater universities when I was teaching and getting the students ⁓ in the company. My boss Andrew Feinberg, the CEO of BostonGene. ⁓ He's a CEO of international company telecommunications called Netcracker. So they had offices all across the world. India, United States, South America, Moscow, Europe, so they had some ⁓ office spaces ⁓ and we have access to the Netcracker brains as as well. David Filby: Wow. and so what's your role and how has it changed throughout those years with BostonGene? Alexander Bagaev: I actually joined very early. I joined as employee number 10. and I joined a company that was seeking the idea how to patients integrate all the complexity of the information about their disease. To be honest, that's kind of a very common idea. When the patient coming to a doctor, sometimes doctors don't recognize ⁓ all the details about the patient's tumor. I'm talking about molecular details, but I'm also talking about all the novel treatment, all the potential treatment options. ⁓ And imagine 10 ago, there was I would say much bigger from what physicians know and what is available. And right now it's even more. We have so many new clinical trials that might help patient. And that was the idea, how we can use the software and novel technologies to integrate all this data together and help select the really right treatment, not the treatment that doctor get used to apply, but the treatment that ideally suits the patient. Well we've been about the true precision medicine. So that was again, that was a very small team brainstorming what to be done. But quite soon we realized that in order to really select the treatment, I mean we're talking about oncology, the tumors are extremely unique per patient. And the diversity is enormous. I'm talking about genetic diversity and phenotypic diversity, plus patient's diversity. So we all combine that into the same story. It means different mutations, different expressions, different structures within the tumor, plus different genetics of the patient itself, different clinical characteristics, treatment history. ⁓ So we quickly realize that we can't only use clinical of the patient. This is not enough. And whatever physicians have in front of their hands, meaning like a basic genetic testing, usually it's like one mutation, few mutations that was approved by insurance, that's not So that's how we pivoted towards developing the precision diagnostic products, focused on whole exome sequencing and whole transatome sequencing. So that's how we started growing and I started getting my first product positions and how I was leading that first genetic product ⁓ where we've been focused on developing bioinformatic algorithms and tools to interpret whole exome sequencing. And back in the days we've been transitioning from the small panels towards the whole exome, whole genome sequencing, how to analyze osmutations, how to apply tumor normal match sequencing And then we quickly realized that genomics is not enough. We need to understand what is happening within tumor tissue. ⁓ There was era of immunotherapies. There was an era of nivolum up, embryolism ⁓ So when we that the immune system actually can fight the tumor. But the immune system, although there is a genomic predisposition to select patients, such as mutational burden. If it's high, you can apply the treatment. it was not enough. Clearly, genetics was not the answer. So we started developing tools to analyze the the transcriptome of the tumor, means RNA expressions and interpreting the tumor micro environment, the immune interactions between tumor and ⁓ host immune system. So we started developing tools to recognize what's happening in tumor through transcriptome. So we pioneered in implementation of the ⁓ sequencing in the clinical among the first we got the first the CLIA, a New York state approval for for the whole transcriptome. and we started working with Md Anderson and bigger institutions to implement this novel diagnostics. So we start growing and growing. David Filby: Wow. And so where is the company now then as we see it and the products that you're bringing to market and that you're developing and your role within that? Alexander Bagaev: So as you heard from the beginning, we've been focusing on developing software tools and technologies to integrate multimodal data together. Now we call multimodal back in the days we called multi-omics, from different angles. And we've been helping select right treatment for a patient. And now we realized and we probably will talk about that from the complexity of the diagnostic space and the insurance, that ⁓ actually we also Develop so many tools to help drug developers select the right patient for a drug. And my view, that's the same problem but solved for big pharma and biotechs, which also lack understanding how to use precision medicine in their clinical trials. Because what what is happening right now usually the drugs are being tested on a random population. patients selected just the smart literature search. Whatever is presented in the publications. to be honest, that's usually that's best guess. we go in for lung cancer because it's a bigger market, or we go into kidney cancer because it's a potentially good market. This is really blind testing. While we know from the beginning that the treatment is not a panacea, it will not work in every patient. So now ⁓ we that we can repurpose tools to help drug developers understand how to make their drugs more successful, means that how to drive the response high. But from the other hand, how to help patients avoid unnecessary treatments, how to select focus the drug development towards the right populations of patients. So nowadays we're building that foundation through the comprehensive AI modeling, we develop the biological foundation models to combine different data from modalities from cancer tissue, patients and immune system. So we're developing the foundation models That's a big AI based infrastructure to analyse this data, understand how the drug works within the continuum of the tumor and immune system, and then how can we design better clinical trials eventually? So that's what we do through the prism of understanding a complex multi level biology of the drug action. David Filby: So the future of the business will have this focus in addition to ⁓ better trial selection, patient enrolment as a way for better outcomes of clinical trials. ⁓ And that a bigger unmet need for you ⁓ or without being of too blunt, is that a bigger commercial opportunity. Alexander Bagaev: That's what we believe right now. The diagnostics space is a little bit cumbersome, especially in US. ⁓ So while we in terms of the science, while we understand all this complexity, while we've been we've been very very successful of our diagnostic product because the physicians loved it. They love the one stop shop. So our product was is that why don't we ⁓ test one mutation? David Filby: Yeah. Alexander Bagaev: Why don't we test just one single parameter? Well, now it's obvious just because that's how reimbursement works. But for patients that's extreme loss. Because when you select the treatment, you need to do a thousand of mutations, you need to know thousands of expressions, you need to know everything about the tumor structure, you need to know everything about the micro environment, you need to know this, this, this, this, this, you need to know that much. So we build this engine and we realise that ⁓ the insurance companies don't want to pay for this complexity. ⁓ was the disconnect. And while it's not solved, we realise that for in terms of the drug development that's a bigger bigger opportunity for us. ⁓ We know how it works and ⁓ can help the patients right away. through the similar technologies, through the similar understanding and through the interpretation of the complex biology where the financially that's more adequate. So there's a lot of money spent on the drug development. billions of dollars are sometimes wasted on drugs that ⁓ might not be a failure. The drugs might be repurposed the right populations and be rescued. And that's what we do. That's what we do right now. David Filby: That's interesting, isn't it? And I think the whole industry knows about the challenges obviously of commercialising and gaining reimbursement for diagnostics. you mentioned joined the company as employee number 10 How big is the company now? Alexander Bagaev: It's it's pretty big. We're about 600 plus people all across multiple countries. ⁓ So I mean I mentioned we office in Houston, Boston, in Europe in Armenia. David Filby: Wow. amazing growth. Huge growth. 600 people around the world. what are the significant milestones that you guys have already achieved outside of the headcount growth? But what's what's the sort of biggest milestones you guys have achieved in the last 10 years? Alexander Bagaev: Of course there's several milestones in terms of investment and recognitions. the first milestone was the implementation of our diagnostic test in US, getting New York State among the first for RNA sequencing, establishing partnerships with ⁓ all major ⁓ And we provide our testing and our test is beloved ⁓ by advanced physicians. I would say the ones that are served for top institutions because we provide that much we provide everything in the same test. not focusing on one particular mutation or event. And of course right now establish the important partnerships with AstraZeneca, ⁓ with J&J and Daiichi, the big pharma companies that we work together and we help them the understanding of the different toxicity effects of their drugs and how we can make the success rate higher. And of course we have a a number of the biotech partnerships with Ottimo Pharma, and many where we help design the clinical trial from from the beginning, from the preclinical data to the early phase one, ⁓ and where we use our diagnostic business ⁓ as for the clinical trials. So it's also the very important branch of us because you know in order to analyze you need to get it. You need to know the garbage in garbage out. So the diagnostic branch helping us to generate high quality sequencing data, the immune profiling ⁓ data for patients. We have all necessary certifications, so we support clinical trials. to do that. ComboMATCH, NCI-MATCH, we also work with the bigger NCI Cohort Consortium to support the new clinical trials that require a novel testing. Novel means not only against single mutation but rather complex transcriptomic ⁓ genotyping. Well we support a SWOG clinical trial for small cell line cancer, which was the important where we start providing the complex transcotomic testing to select the treatments for patients. eventually that could be the companion diagnostics for for this cancer. So I think the important milestones again the initial recognition, initial investments were that helped us to get to the diagnostic market. And then by now we're leveraging our experience, ⁓ the software created and foundation models that we're developing on top of the collected data, to expand our partnerships with Big Pharma and biotechs. David Filby: So you've had huge growth with the diagnostics, diagnostics already and now focusing in addition on clinical trials. Have you had to change how the company is set up and how it will ⁓ succeed moving And I guess the second question on that is are you excited now about the next phase of growth? Because it feels like you're about to open up something potentially you know massive compared to what you've been doing and yet you've grown to 600 people in, 11 or 12 years. It's incredible growth. Alexander Bagaev: ⁓ definitely I to change the team structure and the way how we operate multiple times. Of course initially that it was all about how to establish better operations. So in order to have ⁓ an operations equals the revenues or the better revenues from the diagnostic testing because you need to optimize the the cost per testing you need to optimize the cost per the quality control per report etc. So initially we've been focusing on the automating, automating, automating how to build internal systems, how to build that internal CRM systems flawless. ⁓ you get the and then you get everything in how to use AI to help deselect poor samples, how to perform the quality control using AI. Again, back in the days we've been applying again prediction ⁓ the tissue block is bad, excluded, or ⁓ how can we up the work of the pathologists? So that was the first focus. Nowadays, and of course the focus is on analytics, the dry that's computational side. How and since the beginning we've been a cloud based company. So I think one of our key differentiation from other diagnostic companies, we started not as a lab. I mentioned before Andrew Feinberg, the CEO of the connection to the software cloud world before it's telecommunications. And telecommunications solved this problem many years ago. The problem of how to integrate ⁓ billions of gigabytes of data per David Filby: Mm. Alexander Bagaev: customer or from customers all across the world because again every you generate from your iPhone like ⁓ gigabytes per day. ⁓ And we've been that experience and medicine. From the beginning we started as a software laboratory apply this knowledge to the web part as well. So of course we needed build the laboratory next. but we've been doing that from the prism of technology. So back from the beginning we build the teams that can talk with each other. And it's very complicated ⁓ how mathematicians and software engineers must with biologists and laboratory technicians. And that's nowadays it helps me to expand towards the data science branch and the machine learning branch. I believe we have one of the best people in AI, in diffusion engine, in frameworks that allow us to build from from scratch foundation models to that particular data types. So now now I'm expanding towards the the data science more and more. David Filby: I mean, it's super interesting. What's one of the biggest challenges we hear around diagnostics is sort of education and awareness of the product, right? And it feels like you guys have done that really well throughout your journey because of the partnerships you've established and the growth that you've seen. Is there something that when you look back about what you did really well in order to once you had science and something that worked to ensure that it didn't just stop there, it actually was accessible and adopted. Alexander Bagaev: It's actually a very good point. I say that we succeeded in education. because could have solved the problem of insurance companies and we could have solved the problem of ⁓ of getting reimbursement. maybe that's not your question, but I think education to educate physicians, right? educate the field, those technologies are great to have. And I think when we started working with top academic centers, we didn't need to educate them. They loved the product right away. They wanted the whole exome sequencing, whole transcriptome sequencing, they needed it now. But then we went towards the community hospitals. And here, yes, we faced a huge problem is that why do I need it? How how can I use it? How can I integrate it? How you can simplify it? So we face the problem of interpretation. How to interpret that complex findings and a lot of biomarkers per generated per patients towards simple treat with drug A, treat with drug B, through the prism of the guidelines, etc. And I think that that's what we actually spend a lot of time simplifying the report, simplifying the output, ⁓ using an NLP back in the days, we're using AI. machine and LLM services to kind of narrow down that complexity. So I think still we need to educate the field and I think we've been that along with the other companies. ⁓ again that comprehensive profiling is a must. And of course ⁓ I think the education should shift from from more data to the better interpretation. and ⁓ yeah, I think still an open question. How we better educate the insurance companies ⁓ that this is the way to go. how can we educate the physicians how to use that data complexity. And maybe that's not on them. maybe that's now on us, how to build engines that interpret data and simplifies that complexity for a physician. David Filby: You mentioned you've had some advantage in that your founders of a company had already solved very complex data problems in the telecoms world. Have you also seen a benefit with the AI awareness and adoption of just the regular people over the last few years where it's obviously become really popular and then investment has also flowed into it and you know investors' eyes are all over AI at the moment, have been for a while. But you guys and we speak to loads of people in the industry that have, you know, will say, well we've been using AI for tens of years, you know, it's just now got popular. But has that seen your products become more favorable or maybe easier for people to comprehend and go, do you know what yeah actually that makes sense and that's what we are looking for. Alexander Bagaev: think yes. The the market still needs to understand what the AI is. You're absolutely I'm not sure if again the physician understands what's the difference between different AI tools and still consider to be a black box. Yes, of course we use machine learning before, ⁓ but now we are ⁓ changing our wheels. ⁓ So we are developing David Filby: Yeah. Yeah. Alexander Bagaev: the interpretable AI with developing biological foundation models that allows us to model how the drug works within the patient, how the drug interacts and changes the tumor microenvironment changes the tumor properties in order to see what will happen next. And we train it on top of the patient's data and top of the real data and and we can interpret what is happening. it's a mathematical model is built on very complex math. So in my ⁓ kind of field and what I do, ⁓ I apply that novel technologies, novel mathematics, the novel algorithms to solve the previous problems, how to integrate data together. And just back in the days we used very simple math. We used again machine learning, but that was fairly simple. Now we recognize how to train those systems and create robust models, not overfit models, because that was the fear back in the days. And yeah, let's use the full complexity of data and then we'll train the black box to predict, but it will overfit and it will fail on the next patient. Now it's changing. That's why we're developing that interpretable AI systems. But that's one set. Of course, in an another set of what is happening in the field is ⁓ LLMs, language and how can we use language models to also help interpret patient journey? How can we help physicians leverage BostonGene test, leverage someone else that test linked with ⁓ treatment history that's usually stored in the clinic. How can we all marry it all together through the language models and help physicians treat the patients. I think we still started to kind of educate the field that first of what's this is all possible and second how to control it, how to safely use this data, ⁓ etc. I think we just started. So that's the education just started. David Filby: Yeah, makes complete sense. Yeah. There's still a lot more change needed and I'm there's regulation and so forth to actually allow further adoption in the products. Alexander Bagaev: What I right now is I think both academic centers, pharma companies, ⁓ and companies like us trying do similar things means that ⁓ there is no one have where and who analyze the final data and who should interpret. I think this is the problem or educational problem we're facing is that hospital wants to build internal engines AI engines to interpret data. And they want to get ⁓ the diagnostic results for different testing and integrate it all together. Makes sense. ⁓ sometimes they don't know how to handle this data, how to again because ⁓ data from vendors have batch effects. ⁓ If they come in, they don't have a quite understanding how to deal with batch effects from from terms of data and so we are as a companies, the diagnostics companies, we also build the engines to help physicians select treatments on the data that have. Pharma companies too, they're also developing AI ⁓ language models help to train on internal data. And ⁓ using data, kind of better design the clinical trials. So right I see that three three pillars of the developments that are happening in the field in the end th those three pillars are not connected. David Filby: Yeah, it makes sense. And what happens next then, so for BostonGene, in terms of what's the future plans and milestones and the growth that you're going to see? Alexander Bagaev: So ⁓ our next chapter is leveraging our experience. we have a lot of data collected, which we use for training and as I mentioned, we train our models, and we work with pharma companies to help develop right drugs and right clinical trials. ⁓ And our next milestones to it visible to achieve the specific milestones within our collaborations. So we show how our models help develop drugs eventually and how our models help patients. So a big number of the collaborations happening right now. Some of them we cannot we cannot for some of them we helped rescue the therapeutics For some of them we helped make a decision. This treatment this therapy should be closed. There is no market for this, and we helped our ⁓ customers make that decision. So our huge kind of milestone is make it available to the public. Of course it will help us and it will help the field understand that how real AI, how the biological Models ⁓ can really help. David Filby: Yeah, awesome. And you've mentioned a couple of the challenges in diagnostics and precision medicine. what's your sort of overview on where the industry is with regards to those hurdles and what the future needs to be? Alexander Bagaev: So I think one hurdle as I already mentioned is that the reimbursement is fragmented, ⁓ fragmented. So it means that in the United States we have ⁓ multiple dozens of the insurance companies that everyone requires its own approach. So they have different policies and they require you to basically have a team that works exactly with that particular company. and of course there's no kind of unified approach ⁓ to this. besides that, as I mentioned, the challenge is that even though the scientific field and physicians understand the need for integrated diagnostics, one stop shops. I think it's still not recognized by ⁓ regulatory insurance policies. Because it's just easier to ⁓ approve one measurement. ⁓ so that needs to change, at least in the US. And the last thing is maybe again in the era of AI and you this new models to develop new solutions, we need to understand how to ⁓ work with different data Right right now everyone collects data, we don't share with each other, have ⁓ a minimal understanding what good data, what bad data. So in the field moving moving forward, I think this needs to be discussed and resolved. David Filby: What's your prediction for what happens first? People sharing all the data exists. We just need to put it all in one place. or reimbursement and landscape What do you think's gonna happen first, Sasha? Alexander Bagaev: I think regulation will change first, right now again everyone believes that my data is so unique I will keep it under the lock. They have no understanding why, they have no understanding how to use it. Of course I'm talking about hospitals primarily. ⁓ And maybe I'm talking from the position of technology the software guy, saying, Hey, I know how use your data ⁓ and you don't. Let's work together. you don't need to replicate a BostonGene within the hospital. It's very it's very hard. You will be delaying your in 5 years. Because it will take you only 5 years to understand how to build databases, how to build those engines. In 5 years ⁓ the field will be I guess in space ⁓ compared to where are at David Filby: if you could give advice and I guess the focus on future opportunities and where they are in the industry? You've been with a company who has seen incredible growth. You're still innovating and looking at new ways to problems. But for an early stage diagnostics company or precision medicine company, a company with some really interesting science, where do you think those opportunities are? for them to win and how they should be setting themselves up. Alexander Bagaev: Well, I believe there's multiple opportunities of course. but they to always open what is happening in the field. It's changing. For example the diagnostic testing that was relevant 10 years ago, really relevant now. Well there was a multiple tests that predict immunophere response back in the days But now they don't need it because everyone has been given feminine therapy. So the business idea is So you might have a great test, but you need to know what is the pharma business behind it, what is the actual guidelines around it. Means that they need to be very open to change what testing does. On the other hand, I see a lot of kind of Centralization in the field means that the bigger companies acquire the smaller testing companies. So this is another kind of potential path forward for the smaller companies, but they great diagnostic solutions be to be bought by a bigger player who is ⁓ less agile or you know less smaller companies more faster. so I think this is still a huge avenue for developments and regulations are changing and we better understand science. ⁓ so physicians will recognize the need for the novel testings. ⁓ So always happening. David Filby: I love that Sasha. I think it's really important that early stage companies who believe their science is going to have a very specific impact or be the number one or sole way that something is diagnosed and tested is And I think you're right, if companies are ⁓ open minded and in terms of ultimately how it becomes used and makes a and then as a business you know, generates revenue and survives and thrives, then that's really good advice. Sasha, this has been awesome to learn more about BostonGene. I mean it's a fascinating business. The growth you've seen, the way you guys are doing things and where you're going next super exciting. Thank you so much for coming on the show and helping us to learn more. Alexander Bagaev: Thank you for inviting me. Yeah, that was a pleasure.