theCUBE + NYSE Wired: NYSE Rx | Mary Rozenman, insitro

Clips
More from theCUBE + NYSE Wired: NYSE Rx

Mary Rozenman

CFO/CBO

insitro

play_circle_outline Daphne Koller's Insitro: Machine-Learning-Driven Drug Discovery in a 100,000‑Sq‑Ft Automated Lab

play_circle_outline Overcoming AI's Data Gap: 16 PB Fit-for-Purpose Cellular Imaging for iPSC-Derived Disease Modeling

play_circle_outline From Me-Too to Meaningful: Genetically Validated Targets Boost Drug Success 2.5–5×

play_circle_outline Building a Health Data Moat: Scaling Proprietary Datasets to Enhance Predictive Capabilities, Biomarkers, and Precision Trial Enrollment

Info
Transcript

Mary Rozenman, insitro

Mary Rozenman

CFO/CBO insitro

In this interview from theCUBE + NYSE Wired: AI in Bio 2026, Mary Rozenman, chief financial officer and chief business officer of Insitro, joins theCUBE + NYSE Wired's Gemma Allen to discuss how proprietary biological data is unlocking new frontiers in drug discovery. Rozenman details insitro's automation and robotics infrastructure, which generates about 16 petabytes of cellular imaging data to train a causal AI model of human health and disease. She explains how this approach differs from large language models trained on internet text, since much of biology... Read more

explore Keep Exploring

How does insitro use AI, automation, and proprietary data to discover and develop new therapeutics? add

What exactly is the platform or approach being described, what makes it fit for purpose, and what factors created the “unlock” that enabled its effectiveness? add

Why hasn't the recent influx of capital and AI-driven approaches into drug development led to cures or a significant increase in breakthrough drug approvals? add

What is the third plank of value (after partnering to advance early-stage programs through discovery and advancing programs into the clinic)? add

bolt Powered by CUBE AI

Mary Rozenman, insitro

search

Gemma Allen

>> theCUBE Palo Alto Studio, Silicon Valley and Wall Street. I'm John Furrier, hosting theCUBE here with David Vellante, my co-host. Welcome back to theCUBE Studio here at the New York Stock Exchange. I'm Gemma Allen with NYSE Wired, and today we are talking all things AI in biotech. Joining me now for a conversation on what is shaping the future of this industry is Mary Rozenman, Chief Business Officer and CFO at insitro. Welcome, Mary.

Mary Rozenman

>> Thank you for having me, Gemma. It's a pleasure to be here.

Gemma Allen

>> Those are two big titles you have in your lower third there, Mary, in a very busy and exciting industry too.

Gemma Allen

>> Indeed.

Gemma Allen

>> Maybe just to start, talk to me a little bit about insitro, about the company, your role, and if you can, allude a little bit to this market moment you're in right now.

Mary Rozenman

>> Yeah, so it's an incredible time in our industry, and I think people are rightly more and more excited all the time about the potential for AI to transform many industries, including the discovery and development of therapeutics. So I have had the good fortune of working in this industry for almost 7 years at insitro. insitro is a leading AI-driven drug discovery and development company, founded by machine learning pioneer Daphne Koller. Daphne literally is one of the godmothers of AI. She wrote the book on probabilistic graphical models, was the first AI hire into Stanford's computer science department back in the '90s. And she had this brilliant idea in the early days of the company that, yes, computational approaches were going to scale in this space. But ultimately, over time, the thing that was going to be critical to unlock a new way of understanding and therefore treating disease was going to be scaled and differentiated proprietary data fit for purpose in this industry vertical for these tasks. So what we have built at insitro is, in about 100,000 square feet of automation and robotics lab space in South San Francisco, we have an incredible automation and robotics laboratory infrastructure where we print data in high-throughput agentic automated systems where we model disease in the right cells in which it manifests in the human body. We use assays to produce lots and lots of imaging data. We use AI to analyze those data, and we align that via genetics to multimodal human cohorts. All of that comes together into a causal AI model of human health and disease that allows us to ask and answer a critical question, which is, what would happen in a patient when I intervene at this gene? That kind of insight allows us to develop novel first-in-class genetics-based drugs for ourselves and our multiple pharmaceutical partners. And over time, hopefully expand the predictive capabilities of our model because it gets better every time we turn the crank, every time we use it in a different biologic context, and hopefully bring forward medicines that are going to be really meaningful to patients who don't have great treatments or don't have any treatments today.

Gemma Allen

>> that's certainly a hope we all have right now, right? That drugs that have never— illnesses that have never found a cure will get some sort of reimagination in this world of AI. I want to go back to something you said, and that is fit-for-purpose data.

Gemma Allen

>> That's right.

Gemma Allen

>> What does that mean specifically in this world of biotech? Whose data is that? Is that human patient data? Is that synthetic data? what exactly is that and what makes it fit-for-purpose? And also, Mary, what has created that unlock?

Mary Rozenman

>> Yeah, it's a great question. So maybe just to put things in perspective, there are areas where AI has been deployed where we're incredibly good at making predictions and incredibly have gotten great use out of large language models, for example, in the area of text data. So LLMs are trained kind of on average on about 10 petabytes of large language data from the internet at insitro. In contrast, we have generated about 16 petabytes of data in cellular systems where we model disease in a dish. If we're modeling a neurological disease, we'll use, say, motor neurons or astrocytes or microglia cells in which that disease manifests. For the liver, we'll use hepatocytes and stellate cells, cardiomyocytes for heart disease, adipocytes for studying obesity and adiposity. We print and turn those cells into little microsystems in which we can represent the effects and explore the effects of every single gene in the human genome on that cellular system through a whole stack of modalities. Most importantly, imaging, seeing what those cells look like when you change one of many, many genes. And by doing that over and over again, you can start to put together this massive corpus of data, which is a huge unlock. The power of AI really has to do with what models are trained on. And if you have sufficient data to train the model, the model can learn on those data and do those things, answer questions that are within that distribution. In a way, Well. The problem with a lot of biology is we haven't measured it, and you can't intervene, understand, or treat something that you've never measured at all.

Gemma Allen

>> So let's stay there for a second because you mentioned something off air earlier about motor neuron disease known as ALS in the US and how it's a disease that has never had a solution or any sort of cure, right? Maybe somewhat naively, I always assumed that there were many, many scientists and great minds working in the background around some of these diseases and that just science hadn't yet come to bear, right? But there are probably many cases where, again, like anything, drugs have a commercial element to the model. And there is money spent where money's already been made. What— how true is that thesis? are there scenarios whereby certain conditions or ailments have never really been fully investigated or is my earlier theory the idea that, yeah, they just never really got to the point where tech or science could kind of bring them that extra mile? what's really happening?

Mary Rozenman

>> Yeah. So I think a lot of what you said is correct. So firstly, there are a large number of scientists in this country and all over the world who are working on ALS, amyotrophic lateral sclerosis, or Lou Gehrig's disease. It is a challenge that our entire industry has been trying to tackle for many decades, and it has been intractable to those efforts. So there have been, I think at this point, over 70 failed trials in ALS. I should clarify, 90% of ALS is sporadic, 10% is familial. And for the familial disease, there actually are emerging treatments and treatments that are available to patients that are very, very promising. For 90% of patients, though, with sporadic disease, there really hasn't been anything. So why is that? Right. We've been trying so hard. It takes a global society of innovators to try and unlock this area of disease. what has really enabled our work at insitro in collaboration with great partners at Bristol Myers Squibb is really our laboratory infrastructure, the ability to model, ALS in iPSC-derived or stem cell-derived motor neuron systems. That we can now grow in a way that is extremely tightly controlled through automation, robotics, and stringent controls so that every single step in the process is very, very, very tight and you're not introducing any variability into the system because the system is tightly controlled. We then could use our high-throughput biology platform, which is called POSH, Pooled Optical Screening in Human Cells. We published it in Nature last year. That allows us to go in and then change every single gene potentially in the genome to see whether the cells that look sick, the ALS cells, can go back to looking healthy again. So that's incredible. So AI, imaging, high-throughput biology, all of these tools were critical to get us to this moment where we could ask, is there a gene at which we could intervene to go from a sick cell to a healthy cell? And that's a whole stack of innovations that had to come together. But there was one core one, Gemma, which was that actually we as a society have never been able to see what it looks like for a motor neuron to die from ALS. And by using our AI methods and phenotyping, or taking pictures of cells across that entire spectrum, we were able to build an AI model that learned what it looks like when a motor neuron dies.

Mary Rozenman

>> Wow

Mary Rozenman

>> . And all of a sudden with that unlock, you could see the disease and then you could go in, ask the question, look at all the genes, and finally get to a set of potential genetic targets that have the potential to modulate outcomes in patients. And we say that because we see them modulating those outcomes in cells, we see them modulating those outcomes and causing neurons to regrow in a dish. And we're working very, very hard to move therapeutics forward, both ourselves and with our partners at BMS, that we hope can be important for patients and families.

Gemma Allen

>> But I want to get into your own business model. But before I do, I want to ask you somewhat of a spicy question.

Mary Rozenman

>> Sure.

Gemma Allen

>> It's not targeted at insitro, it's targeted at this industry broadly. Broadly, right? So we know that since 2019, $60 billion has been spent in biotech. it's an area that's definitely getting a lot of focus. And we've heard mixed messages from some folks today about how easily capital is flowing into the industry. There seems to be a mixed view on that. But we know that this is an area of huge possibility. There hasn't been a whole lot of increase, though, in speed for FDA approvals. Right. We haven't seen a huge onset or onslaught of drugs on the market.

Gemma Allen

>> Yeah.

Gemma Allen

>> Why is that? Where is this disconnect? We know AI is capable. It's going to create all sorts of deterministic outcomes, but are we just not there?

Mary Rozenman

>> Yeah. So first of all, I think the fact that so much capital is flowing into the deployment of AI towards better medicines is amazing and is a testament to the fact that all of us as a society want to make each other and ourselves probably healthier. And solve disease and increase healthspan and lifespan for people. So this is an incredibly important undertaking. So I'm really glad that people are financing it. Uh, why aren't we there yet, even though this capital has started flowing into this sector, what is it you said, 5 or 6 years ago? Why haven't we cured all disease?

Gemma Allen

>> Well, more drugs, I guess. More FDA approvals, really.

Mary Rozenman

>> Yeah. So the problem that we have had as a pharmaceutical industry over the past several years is that about 70% of R&D spending is actually allocated towards me-too drugs. Now, what that means is there's already a therapeutic out there addressing the same target, same mechanism, and a me-too program or fast follower program is essentially creating a somewhat better molecule to basically do that same thing, maybe a little bit better. And at the same time as that's been going on, the number of innovative mechanisms that have gone into the clinic has gone down from about 100 per year to 30 per year.

Gemma Allen

>> Wow.

Mary Rozenman

>> Those are not very helpful developments in terms of really driving a step change in this industry, which at its core is really weighed down by the fact that we are just not able to predict early enough in the process whether a given drug is going to be effective and safe in humans. Our average as an industry is between 8 and 10% for how likely a drug is to make it from the very, very early stages of discovery all the way towards commercialization. And so, it actually makes sense. People are going after these MeToo drugs because they're more clinically de-risked and they should have higher probability of success early on. But unfortunately, they're not delivering outsized value to patients in the way that we all should be aspiring to deliver. So what we believe at insitro is that we actually already do know the way as an industry to increase that probability of success. We see anywhere between a 2.5 and a 5-fold improvement in the probability of success of a therapeutic program when that program is going after a genetically validated target. Meaning if instead of just a therapeutic hypothesis that came out in a paper or out of a mouse, we know that that biology is driven by a specific gene in the human body, and we can prove that out in cohort after cohort after cohort on the basis of scaled and statistically powered statistical genetics, those are the targets that are going to have outsized probability of success. So what we have done at insitro is really scaled a massive amount of data and built our entire engine to get to those higher POS targets. That took quite a while to build.

Mary Rozenman

>> I have no doubt.

Mary Rozenman

>> Because the data to do that doesn't exist on the internet. If those data existed, the large language models that are wonderful, both from the incredible Frontier leaders as well as probably some of the open source models would have found those targets already. The data don't exist. This is biology that is vastly out of distribution for what is learnable. So in order to drive those kinds of profound 0-to-1 changes and meaningfully move the needle on probability of success, I think we need to be rethinking how we identify the targets that are worthy of initiating and de-risking therapeutic programs. And the de-risking part is important too, because there are probably programs today in late-stage clinical development that don't actually meet that threshold across the industry.

Gemma Allen

>> Well, let's stay there for a second, because if you can de-risk the investment, de-risk the time, the energy, the hope that is spent on some of these drugs that never make it to market, obviously there's a hugely compelling business case for that amongst large pharma, amongst all of the ancillary ecosystem players. What is the business model for insitro? what is the longer-term game here? Is it around licensing that technology, creating those solutions, again, matching that kind of data to the opportunity in a much tighter way? Or is it also creating drugs yourself? Yeah. Will one thing lead to another? I'm sure it possibly will.

Mary Rozenman

>> Yeah, that's very well said. I think that the notion of one thing leading to another is exactly right. And the interconnectedness between these different components of the business is very much there in how we see the future. So as I've already said, we're building a causal AI model of human health and disease anchored in human genetics. We use those insights to advance first what we believe will be first and hopefully best-in-class therapeutic programs against those targets over time. We're doing that ourselves. We're also doing that with a number of pharmaceutical partners. And over time, as those programs progress, The ones that are owned by our partners will be progressed by them, and that'll be wonderful for patients and for the world. The ones that are progressed by insitro, we will need to decide at every meaningful clinical inflection point for each of our programs. We will need to decide, are we the right people to continue moving that program forward, or is that the moment to hand it off to somebody with a vast global infrastructure for clinical development and ultimately making medicines available to patients?

Gemma Allen

>> So longer-term competitive advantage here, is it around continuing to build out this technology which is novel, which is highly innovative, or is it about the data set that feeds this technology? How do you build the moat here?

Mary Rozenman

>> Yeah, absolutely right. So that's actually the third plank of value. So if the first plank of value is continuing to partner in advancing early-stage programs through discovery, second plank of value is advancing programs into the clinic over time, the third plank of value is very much continuing to scale the foundational data corpus, both through data printing and data scaling through imputation and other AI methods, so that insitro continues to hold a better and always improving causal AI model for human health anchored in genetics. Because that corpus and the predictive capabilities that that affords us and the world as a whole, I think will be incredibly important, not only for uncovering genetic targets, but also for figuring out how to more rapidly enroll patients in trials, how to better track patient state and progressionon therapy, precision health, better biomarkers, all of these types of things could become possible over time if you actually had data that captured biology across scales and across states. If you could say, okay, this is how biology manifests on the level of a tissue, on the level of an organ, on the level of a human, these are the outcomes. And this is how the genetics drives and manifests at every one of those levels. And I can see that on the basis of quantifiable, verifiable data at every scale. That is the future that is achievable if we just continue to advance the platform and continue to scale this data corpus. And we think it holds, incredible potential over time. And in the near term, we are just really excited to be moving our earliest programs forward towards patients.

Gemma Allen

>> And Mary, I have to say, I love your passion, and I love that you said this world as a whole, because sometimes we miss that, right? We focus so much on ourselves and our own lives, but there is a big world out there of folks that need help and need solutions and cures. So I certainly hope you continue to be successful

Mary Rozenman

>> We're working hard

Gemma Allen

>> . Love the female founder story too. Thank you so much for joining us on theCUBE.

Mary Rozenman

>> My pleasure.

Gemma Allen

>> I'm Gemma Allen here at theCUBE studio at the New York Stock Exchange. We are NYSE Wired and today we are covering AI and biotech, talking to some of the key builders and great minds building this industry, not just for today, but for tomorrow. Stay tuned.