Jeff Elton:
Welcome to the ConcertAI Podcast. I am here in the Cambridge Massachusetts studio. Today we are going to be talking about the work we're doing in clinical genomics, translational medicine, and probably some other areas. Mike Rossi is one of our key solution leaders over this particular area. He's been working with all of our molecular partners. He's been working with just about every single one of ConcertAI's customers and clients in the domain.
As this has been coming together, we've been developing a whole range of new methodologies and approach that we think hold real significance for people that are making transition from research phase to first-in-human 1b/2a, and even thinking through strategies from going from one's tumor into alternative tumors. With no further ado, let's start the podcast. So Mike, tell me a little bit about your background and maybe a little bit of pre-ConcertAI and a little bit of coming into ConcertAI.
Mike Rossi:
Thanks, Jeff. Thanks for inviting me. So my background, I have a PhD in genetics and developmental biology. When I finished my undergrad, it was around the time of the human genome project, so I was very interested in genetics and so I just kind of started with cytogenetics. That was the very beginning of genetics and then kind of adapted technologies as I went on. So I did a postdoc at Roswell Park Cancer Institute in Buffalo that was the source of the human genome, and we started doing copy number arrays and I was very interested in the integration of data types, gene expression, copy number.
And then I went on to Yale, and at Yale, that was kind of the birth of NGS. And 454 was in our backyard. There was new technologies developing. The first kind of individual sequence was done by 454 of Watson's genome. And then I went on to Emory as faculty and I was there for nine years and really developed exome, transcriptome, whole genome sequencing, a bunch of clinical sequencing, was involved in something called the Lung Cancer Mutation Consortium. That was really the first incentive to really try to use sequencing to help to direct therapy in lung cancer.
And then in 2017, I left Emory and I went to Mount Sinai and we spun off a company called SEMA4. And the whole goal behind SEMA4 was to really integrate clinical data with the genomics. So what's really important is somebody who comes from cancer genomics world can't really understand genomics of a tumor unless you understand the clinical history. So you really have to interpret what's going on at the genetic level in the context of how the patient's been treated. So whether it's radiation or chemotherapy, really trying to understand that. And then I was lucky enough to come to ConcertAI and I've been here for two years and it's been really a great journey.
Jeff Elton:
Great. And we were fortunate enough to be able to receive you to come here. So in the area of clinical genomics, and this has been a newer area for us. I mean, I think as you know, concert AI really started focusing on the clinical portion, structured, unstructured and beginning to assemble a view of the patient journey and the need to access the unstructured data allowed us to get into some of the molecular diagnostic reports and other aspects which we saw were material to really understanding different attributes of that patient and perhaps why they responded and didn't respond to certain alternative treatments. But there were some limitations.
Then I think as we kind of went in and began to set up this area where we get direct access from the full panel data, which now allows a solid tumor, a full exome transcriptome connected also with this longitudinal clinical record and sometimes with digital pathology and to be included in that. Since this looks like you've dedicated a significant portion of this, as you think about a biopharma starting early stage programs, which really have extraordinarily high failure rates. But as you think about kind of beginning to just sort of plan that first-in-human or perhaps they have a limited data set and they're planning a 1b/2a, how should biopharma today be using these data to inform that study design, either the selection of the target population, maybe even using it to understand who to exclude from a particular trial population, endpoints of interest. How would you think that through?
Mike Rossi:
Yeah, and that's the amazing thing about our data, is really being able to see the patient journey from diagnosis ultimately to death a lot of times. So being able to understand exactly when a pharmaceutical company is trying to identify are they going to try their drug in first line or second line or frontline? A lot of that really depends on having the clinical data to know the patient's been exposed at that certain time point to a specific type of therapy. And then actually having the transcriptome data or exome data to really understand my target is expressed in this tumor after this exposure.
So when you think about what it would cost to get that data, otherwise through a trial, very expensive. $10 million a lot of times to do a phase one trial just to understand is my target expressed or not?
Jeff Elton:
So you're talking about an order of magnitude or more, actually.
Mike Rossi:
Very expensive, very hard to do because you have to go recruit those patients, you have to acquire-
Jeff Elton:
Well, then you have time too, actually, which is-
Mike Rossi:
Exactly. So it's a huge expense and undertaking. And when you think about this, and this is kind of what my career was about, was really trying to understand can we bring sequencing to the patient in real time, use that to identify targets and bring the drug to the patient. What we can do now with real world data is we can essentially run all of these trials and simulations. We can see exactly, is your target expressed.
So when you think about what it takes, if you have an ADC type of mechanism of action, you're going to have to develop an antibody, do immunohistochemistry. There's a lot of contextual understanding that you need to be able to do that. If you have real world data that can simulate all that, you can ask many, many questions across the spectrum, not just your target, maybe a competitor's target as well.
Jeff Elton:
Right. As I'm listening to you, and oftentimes when somebody is putting together a particular trial design, there's internal decisions that need to be made. So it strikes me that some of this is a form of data that can provide more confidence in the internal decisions. Sometimes those decisions, both in terms of the design of the study and kind of even what the intent of the study may be, it may be a decision to then say perhaps this has a lower likelihood of success than they may have thought. Maybe it gets deprioritized, but that's actually a higher value decision to do before you run the study itself.
But it also seems you actually have data here that could actually be part of a informal consultative conversation with a regulator about why are you doing the design the way you're doing the design.
Mike Rossi:
Exactly. And the FDA is always kind of slow to adopt new technologies, new understanding of information, but there's a real need to reduce biopsy burden, limit intervention on the patient's level. And there's now an understanding that real world data quality continues to improve. It's not as gapped, as missing as we had with the early generations of claims data or data that was really from structured EMR that had a lot of missingness. It now has really evolved into something that could be very meaningful to understanding patients in real time right now.
Jeff Elton:
So this is not a question that we talked about discussing in the podcast, but what you're saying prompts me that if I have a 1b/2a, a very early data, and these usually tend to be smaller populations, highly specialized, sometimes the design of these early phase trials also have some branching logic attached to it. You may even get down to smaller numbers if you're following through a branch in that particular study design.
If I'm getting data out at that level, would actually ongoing analysis of the larger patient population where I have that full exome transmission and they have the diversity relative to what these small numbers where I may be now seeing with my agent particular signals and interaction, is there a utility to keeping these analyses going in parallel with each other to help provide and inform even the next phase of the trial itself? Because you're going to go into a 2b/3a in a much larger population at that point.
Mike Rossi:
Very important. And I think for any kind of planning, this ability to understand when to kill a program. My target really isn't expressed very well or it may completely change your strategy. I don't know how pharma could think about proceeding without this level of information because it's now readily available, accessible. The more you understand about patient populations in the real world, the more you're going to be able to figure out your strategy moving forward, especially what may be a no or non go.
When you think about pharma 10 years ago when there was this push to use molecular data to develop small molecules, it was a very different mindset because everybody was very focused on, "I found this novel target, I'm going to make a small molecule." There wasn't a lot of strategy around how prevalent is that target? Do I really have the patient population to support just developing the trial? So having that data accessible and knowing that the data that we have is very relevant and new. So you're not using 10-year-old clinical data, you're using data in the here and now. That's very important for strategy.
Jeff Elton:
Yeah. I know you've been personally interacting with some of our CARAai folks, thinking about doing that word simulation of early stage trial designs. And in that particular area, they're actually creating almost a version of a digital twin of underlying and through an encoder/decoder and creating different attributes to begin to play out different scenarios of response or characteristics.
As you think about the world of generative AI, agentic AI and different AI models that didn't even have reasoning capabilities, which means following a logic to understand why do I predict response, non-response, what do you see the value of these large data sets now up against these future study designs and questions? How's it going to change how we think about working?
Mike Rossi:
It's funny because I always think back to the early days of IBM with the whole GIGO acronym, garbage in, garbage out. I used to use this when I lectured, this idea of when you sequence a tissue biopsy, you actually have to know the quality of that tissue to know what kind of data you're going to get out of it. Even more so now with AI. So if you have poor foundational data, whatever AI is capable of is going to be erroneous because it's not really getting you to the right answer.
And it can take you down some paths that you might not want to go down. Incredibly powerful tool, and I think agentic AI is really going to get us to the point where there's even more accessibility. When you think about just the ability to interact with code and change parameters as you're going, as knowledge is gained, I think that's really going to be transformative in medicine. And that's always been the goal. The goal was always, can you link the data with technology to advance and speed innovation? And I think that that's really where we are now.
Jeff Elton:
Yeah, the whole GIGO thing, I remember, and I think I used it quote a bit at the time too, but I think one is the data substrate itself has to be considered the right data for the classic question and decision you're going to be making.
Mike Rossi:
Exactly.
Jeff Elton:
Which both may have to do with scale, representativeness, characteristics, assuring it actually was the right portion of the tumor material, so you know you had the right cancer cells and you're getting the right view of that particular patient. The second part of it is these tools themselves actually will become adjuncts of highly skilled individuals collapsing some of the time of their work, not really taking away the expertise needed in individual effect. If anything, maybe you need the expertise in the individual, but what might have taken months or years can collapse down to days or limited number of weeks.
Mike Rossi:
Exactly. And I used to always have this conversation with pathologists was this idea of the integrated report. So if you think about the medical record, it's a concatenation of reports. One test after another, and it was never synthesized in a way that made it readily accessible to a physician.
I feel we are now entering this phase of medicine where the data is now there and we're going to be able to present it in ways that a true expert's going to be able to look at that and synthesize and make a lot of decisions. And I do not think we're at the terminator stage where it's going to take over. I think there's still a lot of human control required to understand what to do with it, but I think it's amazingly powerful.
Jeff Elton:
Yeah, so much of what we've been describing would be acquiring tumor material from, and we're kind of looking at the genetics of the tumor itself, obviously not, but there are now liquid tumor biopsies which are actually using serum, and there may be multiple time points. Sometimes this is used for high-risk surveillance as well as over the course of a treatment and MRD and minimum residual disease characteristics, et cetera. How do you see liquid and solid kind of beginning to play together from a research perspective? I can understand from a care perspective, but how do you see it coming together from a research perspective?
Mike Rossi:
The thing is that we can do more and more with the liquid component, and the plasma component, the CT DNA, even circulating protein and RNA, much more accessible than a biopsy. So a lot easier to collect blood from a patient. Because the technologies have continued to improve, the sequencing technologies have continued to improve, we can now make assessments on very, very small, it's really at the molecule level. So being able to understand a patient before a specific intervention and then after an intervention, much easier to do with blood than with tissue. And it's really just a question of those technologies, they call them a minimal residual disease type tests, highly specific, can really identify very low levels of disease really starting to transform the way we understand response.
So if you think traditionally how response is assessed in a solid tumor, it really requires radiology, and this is where multimodal becomes so important because it's going to be a combination of digital pathology, radiology, tissue sequencing, blood sequencing, other types of modalities that we aren't even really thinking about and integrating. Things as kind of standard as a CBC blood profile, integrating that into the larger collective data set. We're going to really start to triangulate patients who are and aren't responding very quickly and being able to intervene much sooner in terms of their care.
Jeff Elton:
So what I'm hearing from you is I may find different views of what response in different cancers look like with different treatments, that might even inform also parts of my trial design because I know what I'm measuring to. And I actually may also be able to use liquid kind of as a measurement system over the course of the trial, including maybe discontinuation at a point if I'm not seeing response or kind of things of that particular nature. So I can see it playing into the whole clinical development process in an array of different ways that are out there.
Mike Rossi:
Right.
Jeff Elton:
Now that you can almost predict which patients may have acquired resistance and then therefore may be more responsive and actually develop algorithms to select patients that actually for a new treatment that have the highest likelihood of being a high responder. Do you see those almost, whether it's being done as software as a medical device, or do you see those as tools? With all the use of AI and algorithmic approaches that now can be deployed in live electronic medical record environment, do you see bringing that information and knowledge package together as a model with a therapeutic as being a way things might evolve?
Mike Rossi:
I hope so. I hope so. So it's always the question of medicine never moves at the speed that you want it to.
Jeff Elton:
For sure.
Mike Rossi:
And there are always early adopters in any sphere. Being very fortunate to have had the academic career that I had, I was able to see physicians that were on the cutting edge. They were pushing every therapy. So in multiple myeloma, in lung cancer, I saw oncologists that were doing everything for their patients and thinking three steps ahead. There are always other domains within medicine that are really just trying to put the brakes on things and move things slower.
So there will be an early adopter phase that I think is happening right now and the really progressive physicians are going to test the waters. They're always going to be the naysayers who say, that failed, it didn't work. It's going to be something that we're going to have successes and failures. We're probably going to inch our way along. But I do feel like we're going to hit a growth phase where it becomes much more readily accessible. I love the idea of somebody at a rural community hospital getting the same standard of care as [inaudible 00:18:42] Sloan Kettering.
Jeff Elton:
Absolutely. But at the end of the day, the biopharma innovator probably actually has the most at stake to make sure-
Mike Rossi:
Exactly.
Jeff Elton:
... that that drug gets to the right patient that's actually going to get to the right outcome. And while we're getting intelligent infrastructure and models and things that can be deployed in clinical augmented decision tools, they all operate at a very high level, probably not necessarily the individual therapeutic. So it'll be interesting to see how this evolves.
Mike Rossi:
Yeah. And it's just going to make the space that much more competitive, right? So it's going to require pharma to be more innovative, think about more novel combinations, work together to try to find... Rather than everybody looking after the same target, really trying to figure out how you position your drug.
Jeff Elton:
Right. So let me ask you one last question.
Mike Rossi:
Sure.
Jeff Elton:
So as you look out over the next two years, sort of for the back half of this year and going into '26, what are you most excited about? What gets you going in the morning and kind of wanting to work with all of our biopharma partners, et cetera?
Mike Rossi:
That's a very tough question because it's very broad.
Jeff Elton:
Okay. You can open up the aperture of your response a little bit on that one.
Mike Rossi:
So on the technology side, amazing what we're doing in the sequencing realm, spatial pathology, single cell. A lot of that data now, being able to understand what every cell in the body is doing biologically is incredibly fascinating to me. That's where a lot of the CARAai stuff comes in, understanding components, cellular components of the tumor as well as the bulk tumor, really amazing.
I think our pharma clients are really, they're breaking down into tiers and so there are really innovative companies that are out there really trying to push the envelope. A lot of people that we work with on a daily basis, very excited to see how they're designing trials, which is really, they're using real world data now more than ever. And two years ago it was predominantly still in the realm of outcomes. Being able to use that data sooner, feedback into pipelines, I think that's the most exciting thing to think, they're going to learn from a trial, we're going to use the real world data.
Jeff Elton:
I think it's huge. And actually, if we can even do continuous measurement of the data in the trial itself, I think we can take all aspects of this and speed the entire thing up. And if we start actually with deeper insights and deeper purpose and narrower sphere of questions we're looking at, that actually will actually change the success characteristics of those early stage trials as well.
Well, Mike Rossi, thanks again for being here in Cambridge, visiting the podcast studio.
Mike Rossi:
Thanks. Thanks so much.
Jeff Elton:
Always a pleasure. Always learn something every single time we sit down and talk about this. So to everybody who is listening, thank you very much as well, wherever you are. Good morning, good afternoon, good night, and we'll talk to you next time.