Decoding the Immune System to Optimize Immunotherapies

Daniel Levine (00:00)

The Life Sciences DNA podcast is sponsored by Agilisium Labs, a collaborative space where Agilisium works with its clients ranging from early stage biotechs to pharmaceutical giants to co-develop and incubate POCs, products and solutions that improve patient outcomes and accelerate the development of therapies to the market. To learn how Agilisium Labs can use the power of its generative AI for life sciences analytics, of viral infections or bacterial infections. And if it is hyperactive, you can start developing autoimmune indications where the body attacks itself. Now the immune system is a very complex system where you have literally trillions of immune cells that are in your body that are generated in different organs and they are being kind of pulled in different organs, they start in the thymus, the T -cells, or the spleen, and different organs are being really primed in the lymph nodes. And it's a very complex system where immune cells are affecting one another and affecting the organs that they are being launched at, and they communicate by sending cytokines or chemokines. And think about...

how a very complex system of trillions of cells that is inside your body is interacting with one cell and another. And being able to measure this immune system is one of the very difficult open problems in biology and immunology. And when we started back in 2018, we had a hope that by leveraging...

really hardcore engineering capabilities, single cell technologies, different types of measurements of tissues. We can measure cells within tissues. To help you turn your visionary ideas into realities, visit them at labs .agilisium .com.

You're tuned to Life Sciences DNA with Dr. Amar Drawid.

Amar, we've got Noam Solomon on the show today. For audience members not familiar with Noam, who is he? Noam is a co -founder and CEO of Immunai. He's worked both within the industry and academia. He has a double PhD in math and computer science and served as a postdoctoral researcher in the mathematics department at MIT and in the Center of Mathematical Sciences and Applications at Harvard.

And what's Immunai seeking to do? Immunai is combining immunology with machine learning to develop the next generation of immunomodulatory drugs. It's applying that technology to discover and validate targets, validate compounds, and optimize clinical trials. And what are you hoping to hear from Noam today? I'd like to understand what's unique about Immunai's approach, the data it's able to amass, and what kinds of novel insights

It’s able to extract from this data.

And before we begin, I just want to remind listeners that they can subscribe if they like our content, hit the like button, and also remind people that if they want to hear an audio only version of this podcast, they can find it on most major podcast engines. With that, let's welcome Noam to the show. Noam, thanks for joining us. We're going to talk today about

How Immunai is leveraging AI to develop next generation immunomodulators. Before we talk about Immunai technology, I want to start with the company's founding. Both you and your co -founder, Luis Voloch, had been trained in machine learning, but you both had experienced a need for better cancer therapies through family members. Can you explain the role this has played in the founding of the company?

Yeah, definitely. And thanks for the invitation to be here and speak with you. So the company was founded back in 2018 when I was doing my postdoc research at MIT and Harvard. So I was living in Cambridge at the time. And the problem that my co -founder had was a family issue. His grandfather, unfortunately, had cancer and he was getting a combination of immunotherapies, which are immunomodulating agents that are supposed to get the cancer to be in remission. And this was indeed the case, but the side effects of this combination treatment were too difficult, too severe for him, and he decided to stop taking the medication. So this was back in 2018, not too long after he unfortunately passed away. But this was for me and for the company, the founding story. We really wanted to understand how can we leverage data to guide better decisions, better precision medicine decisions for patients. Okay. And how did you connect with Ansu Satpathy and Danny Wells? And what was the significance of the work that they were doing?

So roughly a few months after Luis and I decided to found Immunai, we started approaching different researchers in the space of immunotherapy, pathology, oncology. The first one was Ansu Satpathy, he was a young researcher at Stanford University, principal investigator. And Ansu's research was about studying the effect that...

immune or modulating agents were having on the immune system by leveraging cutting -edge technologies at the time called single cell technologies. And by using those technologies, he was able to find insights that were new at the time on how these immune checkpoint blockade therapies were affecting the immune system that you didn't know before. And Danny Wells was a collaborator of Ansu, so he also joined the founding team and we founded the company.

Together they joined the scientific co -founders. They kept their day job so to speak and we launched it full -time, Luis and I. So before we get into what Immunai does, can you talk about immunomodulation? Because that's a pretty big word. So people who are not familiar with what that means and in general how immunology works, can you talk a little bit about that?

Yeah, definitely. So maybe we'll mention the immune system, which is kind of the fundamental system that every one of us has. And the immune system, when it operates well, we are all healthy and we remain in good health. But when it starts going sideways, we develop different types of diseases. It can be, unfortunately, cancer if the immune system is not effective enough, or you start developing different type  cell resolution. And creating these data sets, uploading the data sets, in our case to the cloud, creating more and and more data sets of the immune systems of patients that are being treated with immune modulating agents.

Maybe by seeing how certain patients are responding to the same drug, the same immunomodulating agent, and some of them get better, whereas others are getting the same therapy, but they are not getting better. They are, you know, progressing, they are progressing and they eventually die. Maybe by being able to understand the differences of this immune system of the different patients, we will be able to decode the complexity and be able to guide better decisions. So immune modulation is what happens when a patient is receiving a drug. The drug is supposed to impact their immune system. And it is a little bit like a domino effect where one, you know, the beginning is binding to the target. And then one immune cell is affecting the next immune cells and next immune cells and being triggered and being deployed in certain organs.

And eventually, if the drug is successful, you're going to see a good immune response and then a clinical response. And sometimes you don't see the good clinical response. So when you're talking about the immune cells, you have a lot of different types of cells, right? These are the white blood cells. So you have T cells, B cells, you have macrophages, neutrophils, right? So a lot of these different types of cells, and they are coordinating with each other. As you said, they are sending signals through molecules like cytokines and coordinating with each other. But what you're saying is that for a specific disease or so, different patients are responding differently. And is that what you're trying to measure or are you talking about like, of course, like when you have different types of disease, the immune response will be different. But then also for the same type of disease, the different types in patients, the differences will be there. Is that what you're trying to say?

Yeah. So indeed in the immune system, in your immune system, in my immune system, you have different types of immune cells. You have T cells, you have B cells, you have macrophages. And it's not just that, different T cells also exist. You have CD4 cells that are called the helper cells. You have the CD8 cells that are the killer cells. And within each category, it's like a decision tree that you go down the tree and then, you know, CD4, you have TH1 cells, you have TH2 cells, you have TH17 cells.

And they also have different cell state activation. And by being able to say in certain patients that both have T cells, but the abundance of the T cells or the abundance of subtypes of T cells is different, or maybe the genetic makeup of specific subtypes of T cells or B cells or macrophages, are the reason, the underlying reason why the immune system of these patients do not respond to the drug.

And other patients that have different types of those B cells or T cell abundances or genetic, more precisely, more genomic expression of these cells is different. And because of this difference, you're going to see a good response or a bad response. Now, if it is one patient versus another patient, there is very little you can say. But if you're measuring hundreds and thousands of patients and maybe more than that, then maybe you can find the common denominator for good response and the common denominator for bad response, resistance. And by being able to measure more and more of these immune profiles with more and more granularity, you may be able to kind of figure out the reason. Okay, gotcha. And the interesting thing here is that you have like very different diseases like rheumatoid arthritis or like asthma or cancer and...

Here we're talking about immunomodulation to treat all these really different types of diseases. When people understand cancer, they're like, well, these are some really bad cells which are reproducing like crazy. Can you talk a bit about how can immunomodulation on the one hand affect rheumatoid arthritis, on the other hand, asthma, but also on the other hand, cancer? Can you talk a little bit about that and how...

immunology can be programmed in different ways, right, to attack different types of diseases.

Right, so that's indeed the fundamental question. We are given one immune system and we need to use this immune system to cope with everything. You know, unfortunately, when people, you know, grow up and they grow old and eventually they start having different diseases, I think most of us at some point had COVID, right, a few years ago. Some of us or many of us were fortunate enough to have very mild symptoms, maybe even completely be asymptomatic, whereas others had severe...

effects, and some of us even died. The thing is that the same immune system with the same cells have to cope with different diseases. And like you said, cancer happens when your immune system is starting to not identify bad cells. And those cells are cells that produce proteins that should be detected by the human immune system. But somehow, it isn't or they aren't. And because they aren't, they start reproducing. And if the cancer progresses in our body, it means that the immune system is kind of blind to its role, which is to identify, attack, and destroy these bad cells. On the other side of the spectrum, the immune system also identifies sometimes the wrong types of cells. It starts affecting ourself. And so it starts affecting...

good cells that should be there and without harm, but somehow the immune system decides that they are foreign cells and should be attacked and destroyed, which is the opposite problem. In one case, the immune system is not doing enough to identify bad cells. In the other case, it is doing too much and it identifies good cells as bad cells.

And so in one case you're going to get cancer, in the other case you're going to get autoimmunity like RA that you mentioned and other indications. And there are other examples of other diseases. In fact, even aging is an immune issue. We age because our cells age and there is something about the immune system. If the immune system is better, you may age less quickly. And if it is not as strong, you may age faster. So the immune age of people is a very interesting concept.

Well again, you may be able to reduce how fast you age because the immune system is stronger, the immune cells are doing better. Interesting. Yeah. And the main function of the immune system is to kill off the viruses and bacteria, right? That's what it's built. And so basically what you're saying is that that's the original function and it's basically killing the viruses, bacteria. But then similar to that is the cancer because cancer is these bad cells which are growing like crazy and so the immune system needs to kill them off as well, which it's not doing. But then sometimes we have the immune, so in these diseases like the bacterial or viral infection or cancer, the immune system needs to do more, right? Needs to be more effective to kill off those germs or those cancer cells. On the other hand, there are times like rheumatoid arthritis when immune system is attacking our own cells for no reason, so it's become hyperactive and it needs to calm down, right? So basically what you're saying is that with immunomodulation, you have to basically, sometimes you have to increase the effect, sometimes you have to decrease the effect, sometimes you have to change the effect, right? And that's what you call immunomodulation, is that correct?

Correct. So the immune system is responsible for keeping good law and order to keep the equilibrium. And so in some cases, you can think about it like the police or like the army that you want to keep good law and order. So if some force is behaving with bad intent, you want to kind of take it out and you also want to keep good people safe. You don't want to, God forbid, to attack the good righteous people. And I think this is exactly the same with our body. You want to keep the intrusions out

and you want to keep the good cells without creating any harm for them. So that's kind of the role of the human immune system. And also for other animals, almost all of them have immune systems. Great, great. And so in terms of the immunomodulation or immune therapies for cancer, we've heard a lot about Keytruda, Opdivo, the PD -1 inhibitors, right? So can you talk a little bit about that? And also can you talk about...

the future of immunotherapies against cancer. So how that whole area has evolved over the last few years and how you see it evolving in the future.

Right, so there is an area in immunotherapy that is called immune checkpoint blockade therapy, which are essentially the ability to unblock the good cells from being activated, for example, to kill cancer cells. And one of the best examples, I should say the best example out there is PD -1 therapy, most, you know, best known in Keyruda or Opdivo. And it is, I think, the best selling drug out there for cancer.

There are other therapies that also have been approved. They are also effective, but they are not as effective. And I would say that many pharma and biotech companies are looking for the next Keytruda. That's kind of the race to find more therapies. It is probably true that we are not going to find a drug exactly like Keytruda, a drug that has transformed precision medicine, especially for...

melanoma and non small cell lung cancer, but also in other indications. In many indications, it is given as a combination therapy, so it is going to be given in combination with other therapies. And as the CEO of Immunai, we are looking for more ways to find good therapy that are going to improve patient outcome as much as Keytruda and Opdivo change the space. Now, the future of immunotherapies

is very, very, very promising. There are multiple ways to go after novel immunotherapies, but maybe I will say that the concept of immunotherapy is the concept of developing drug that are targeting the immune system. They're not targeting the cancer or the autoimmunity. They're not targeting the diseased tissue. They're targeting the immune system. They're trying to empower it and make it stronger with better equilibrium

because the understanding is that if you target the immune system and you make it stronger, you're also going to be healthier. And there are different ways to go after the next generation of immunotherapies. You can go after cell and gene therapies. So these are engineered cells that you take sometimes from the actual human body of the diseased patient. You're taking it out, you're trying to engineer it, and then give it back to the patients.

In other cases, you're trying to do it outside the body to find good donors to treat patients by actually engineering those donor cells and then giving them to a wide variety of patients. There is a dream to do it, you know, an off -the -shelf cell therapy that is going to be a much wider industrial application of this thing of giving cells back to patients. It's a more complex mission.

It's not like giving an antibody or a small molecule. There are, as we discussed, antibodies like checkpoint blockade therapies that are still in pursuit. There are are many companies going after them. I should say that except for oncology, where in solid tissues, solid tumors like melanoma and non -small cell lung cancer, I think some of the best use cases are in hematology. In blood cancers, we see a lot of...

good outcomes for patients. But I think that in many cancer indications, probably 75, 80 % of patients are still not treated with immunotherapies. And so the promise is huge. And the number of applications, whether you go after siRNA, mRNA therapies, cancer vaccines, you want to go after new ways to deliver drugs, nanoparticles, et cetera.

There are so many great ideas out there and I think the future is very promising in terms of new immunotherapies. So, Immunai started out by building the atlas of human immune systems. So this, I believe you call it AMICA, the acronym for Annotated Multi-Omic Immune Cell Atlas. It's using patient-derived multi-omic data to drive target and therapeutic discovery. So, can you explain that a bit and where does the data come from? What does it include?

Yeah, definitely. AMICA, as you said, is our proprietary database and knowledge base where we've been curating for almost six years now different types of immune profiles from patients. But maybe I will say we talked a little bit about the immune system, about immune modulation, about immunotherapies. But I want to also discuss how do you profile the immune system? We mentioned T cells, B cells, macrophages, dendritic cells.

But how do you profile this? There isn't a canonical way. And what we've been doing in the first few years of founding the company was to figure out a canonical way to measure the immune system of patients. We take a few thousands of cells, of immune cells, from patients, usually both from the peripheral blood, which are the white blood cells in your blood, and also from diseased tissues. Let's say in oncology and cancer, you take biopsies from...

tissue and we measure the immune cells within these tissues. We measure those immune cells with single cell technologies. So we measure the RNA, the mRNA molecules in the cells, we measure surface proteins, we measure the T -cell receptors, we measure different omics. And for each of these cells we do it, we don't do it at bulk. So you get a very large matrix that measures the immune system of a patient.

So think about this as measuring 5 ,000 cells and for each of the cells you measure, let's say 20 ,000 different genes, 200 different surface proteins, the T cell receptor. It's a very high dimensional matrix that we measure just for one patient sample. We try to measure for each patient different time points. So we are measuring these metrics for patients before and after therapies, usually different time points after a therapy. And then we try to measure,

large cohorts of patients being treated with the same drug. And we do it for multiple drugs, for multiple indications. And AMICA is the collection of all the data sets we are building together. So a lot of the data came by us working with academic collaborators and with pharma and biotech companies doing clinical trials. And our platform is being used to map the human immune system of patients before and after the drugs that are being administered to them.

We are growing the database. Today we have a little bit over 10 ,000 samples in AMICA with single cell resolution. And the goal is to get to a million samples. So that's kind of the goal that we have going forward. So to double click on this is for one patient, how many different cells measurements are you doing? So how many different cells? Because then for each cell, I understand,

for each measurement you're measuring 20 ,000 genes, but then you're doing probably genomics, trans proteomics, proteomics, and then you have multiple cell points. So it's like all multiplying, but like how many different cells or how many different cell types, can you talk a bit about that?

Right. So we usually, the golden number that we use is 5 ,000 cells per patient. And it's an optimization problem of the measurement and the cost. Because if you're going to take 50 ,000 cells per patient sample, it's going to be 10 times more expensive. And in terms of mathematics or the information of the problem, we find that 5 ,000 cells is very good to contain most of the information of

a human. And you're measuring different types of cells, like dendritic cells and like different types of T cells and stuff or? Right, so we are measuring them in peripheral blood. We are measuring all the cells that we can find kind of uniformly at random and we also measure from the tumor tissues. So there is an issue that some cells have better, you know,

attributes that allow them to survive longer. So the distribution is not going to be identical to how they are in the blood, in the body, or how they are in the actual tissue. But over time, we get a fairly good distribution of the cells. So you're going to get dendritic cells, you're going to get T cells, B cells, and different types. And the important thing is that it's not just kept as a T cell. For a T cell, we're going to measure 20 ,000 different genes, hundreds of surface proteins, and we're going to be able to go...

in this decision tree that I described, very deep into the different subtypes and sub -states. So in a T cell, we're going to see it's a CD4, Th17, and then the specific activation or cell state of the T cell. And with this, we're going to be much smarter in analyzing differentially expressed genes or proteins between, let's say, responders and non -responders. Okay. Okay. But then just doing the math here, you have per patient you have 5 ,000 cells,

20 ,000 genes in each cell, you just multiply that, that gives 100 million. And then you have different types of omics and you have different time points. So we're talking about for each patient over a billion data points. Is that right?

Yeah, you're getting very high numbers and it depends on how much you, how many measurements you want per patient. You can get even a hundred times more than you just mentioned. The important thing is that it took us a while to optimize the magical number of how many cells we believe we need in order to kind of compress the immune profile of the patient in order to get meaningful insights. But yes, for each patient sample, you get many dimensions. Okay.

And how are you even storing all of them? This is a huge amount of data.

Right. So that's part of the beauty and appeal that we have to a provider, cloud provider, they all want to work with us and over time becomes even more and more excessive. And I agree with you. This is why I think that Immunai is an engineering first company because the fundamental challenge is, okay, you have the technology to profile the human immune system or the immune system, but then you...

start generating data and the data becomes very big data and now you need to start running algorithms on the database. It becomes very expensive, important for you to have very accurate and relevant infrastructure tools to actually do it efficiently because the prices and the cost can be very high if you don't do it well and they can also be very timely if you don't optimize. You don't want to run algorithms for years. You want to be very efficient.

And so you're applying machine learning to this data set. So can you tell us about what type of novel insights you have been able to glean using those algorithms about the immune system?

Right, so there is a variety of insights and maybe I'll start by giving a few examples. I think that the most basic examples are like scientific insights. So when you analyze the difference between, let's say, patients and healthy people, patients that have specific, let's say, non -small cell lung cancer and healthy people, you are able to say, okay, I see some immunological insight that I think is very relevant or very interesting or very important.

There are also translational insights. The translational insights are insights that are connecting the science to the clinical world, right? So you can say, I think that in specific types of cells, let's say Tregs, that are also called regulatory T cells, in those regulatory T cells, a specific expression of a specific protein or gene is highly correlated with a good prognosis. And I want to monitor this

in many, many, many patients with this indication being treated with the drug, for example. The last example is things that are driving clinical value or business value. And this can be about how do I optimize a clinical trial? How do I find the optimal combination agent? So how do I find the best chemotherapy to give the drug that I'm developing because I want the FDA to approve it and it's not enough to have mildly better results. I want to have much better.

So I need to find the best combination agent or what is the best dose to give the drug to the patients and what are the right indications to go after because the same immunotherapy can be given for a variety of different disease indications. Each of these questions, if answered properly, can really optimize for the success of the clinical trial. And maybe for our viewers and listeners, the process of developing a novel immunotherapy,

can take more than 10 years, sometimes 15 years, can cost more than $2 billion. In oncology, more than 95 % of investments fail. So even after you have a drug candidate, you start giving the drug in phase 1, phase 1b , phase 2, it doesn't get the FDA approval. You spend 10 plus years, you spend $2 billion, it doesn't make the cut. And so by leveraging...

a technology like we developed, there are use cases about choosing the optimal dose for the drug, or choosing the optimal chemotherapy to give the drug in combination with, or choosing the optimal indication to go after. These are examples for case studies that we've performed successfully with our pharma partners that have demonstrated the strength of being really able to map or decode the human immune system and the differences between responders and non -responders to the drug.

Or between the patients that have severe adverse events and being able to decode or unlock toxicity. These are questions of critical business value. And these, like the clinical questions that you talked about, right? How much you're seeing, like as you are profiling these cells at a, you know, such granularity. Are you seeing, have you started seeing like clear connections between the clinical manifestation

and the genetic profiling that you're finding with the cells?

Yes, absolutely. So I think that in our case, we have developed an engine that is called the IDE, the Immunodynamics Engine, where we are studying for all the data that we have. We are studying the dynamics the drug has within the immune system. And you can think about this as a drug being given to hundreds of patients,

and you are trying to separate between the responders and non -responders, or between different dose levels, or between different clinical arms. And what the algorithm does is a very sophisticated matrix factorization. At the end of the process, the algorithm finds the top rows or the top features that are correlated with the outcome. So the top feature can be a specific cytokine that is very relevant to the effect of the drug,

or it can be the abundance of specific subpopulations of cells. So it's not just how many T cells I have, but how many very distinct subpopulations of cells is in the tumor or in the blood and using this as a proxy to tell you that this is going to be the right dose to go after. And so it's not just to tell you that we are doing...

analysis of the clinical data. No, it's molecular profiling of the immune system. And then the outcome is an insight that is an insight about your immune system with very high granularity that is correlated with the question you're trying to solve. And that's the outcome of our product or our platform. I think that the challenge that we have, and I think the challenge is across the entire industry, is that there are so many...

different immunotherapies are being given and they're being given to very small cohorts of patients. So I think the challenge is to be able to really trust the platform and build the AMICA to be larger and larger and larger. As I mentioned today, it's about 10 ,000 samples. We want it to be even 100 times larger. In every scale of the data, there is a business value that can be derived out.

And we are in a very exciting stage. We're already in a 10 ,000 regime. We are producing valuable, actionable insights that can drive business value for our partners. And the kind of machine learning algorithms you use, are those more of the traditional machine learning algorithms? Are you using neural networks? Can you give us some idea about that?

Definitely. So we are using a variety of tools and maybe I'll mention a few technical notions. I think in the space of multiomics where you're trying to use single cell technologies, for example, single cell RNA sequencing is a very noisy technology. So you're measuring the mRNA molecules within cells, but the tool that is measuring the mRNA molecules is very noisy.

And also RNA is something that degrades and degenerates really quickly. So you take a biological sample out of a human patient and the cells start dying really quickly. It's not like DNA, the DNA can survive for a long time, you know this, but the RNA degenerates really quickly and start dying really quickly. So the processing wait time that you have between the time that you took the biological sample out of the human and you sequenced it is very important for the results that you're going to get. And when I mentioned that we do immune profiling of the samples, a lot of the novelty in machine learning and AI is about representation of the RNA data in a structure that really corrects for the batch effects and the noise. So this is a very important part of the technology.

And the way that you do it, you use different types of methods, including deep learning algorithms that include auto encoders and can also include the new transformer based algorithms that we are all familiar from, you know, OpenAI or DeepMind, we are using these algorithms and they are not applied like a cookie cutter, one size fits all. It depends on the question. In order to apply those very sophisticated models, you need to have a lot of data

to control for the very high dimensionality of the problem. You said earlier that we measure a lot of measurements from a single patient and the number of dimensions is very high because it's the number of cells and with each cell you have tens of thousands of measurements. So you really need to have a lot of data to create these very fancy deep learning models. The way that you can...

create deep learning models on smaller data sets is really as much of an art form as it is science. And I think we are all dealing with trying to create the ChatGPT for single cell data. And that's what we're also doing within Immunai. But I would definitely say that we are leveraging classical machine learning together with deep learning, not one or the other. And it's related to the specific problems we are trying to solve.

It requires a lot of sophistication. Half of our company are computational people. We are today 170 people across four different sites. Half of the company is computational. We have a lot of computational biology people that are trained in biology and immunology. We have machine learning and AI people that are trained more on the mathematics and physics and the AI. And they're working together to try to figure out what are the best implementation and algorithms to develop that are going to tackle the problem. problem.

So it's not a one size fits all, transformer based, large language model is going to be best for every problem that you're going to give it. So the company has taken a lot of steps, both in terms of partnerships and acquisitions to expand the data. So what have you done to expand the data set today?

I mean, it's hard to get all these samples and all these patients, right? So then we talked a lot about like the biology, we talked a lot about the technology. So now it's just about, okay, well, how are you working with the environment, right? Like to get the more data sets.

Right, so I started saying that we are profiling many human patients from our collaborators, from academic institutions, from pharma and biotechnology companies, but this is not the only way that we are growing the data sets. We are also doing a lot of perturbations or immuno-modulation in the lab. We're doing this in vitro, ex vivo and in vivo, so that means that we are taking lab models and we are testing what happens when you are modulating specific...

immune cells or immune systems variants. We're also doing work in vivo, which means that we are doing experiments in different types of mice. And those are very important aspects of how we're growing AMICA. Because drugs work in mysterious ways, especially immunotherapies, we don't really understand how even Keytruda, more than a decade after it is being developed, how it is working for patients with cancer, with

ovarian cancer. We don't. We know some parts of how it works. And we want to understand better. So the question is, can we develop lab models that are going to shed light and improve our capability to predict what will happen to a pancreatic cancer patient or ovarian cancer patient? And the answer is that in order to do it, you need to have a database that is going to have for the same drug, let's say, Keytruda,

the Keytruda being given to patients with different types of cancer indications, but also try it in different types of mice, humanized mice, and genetic mice, and different types of other preclinical ex vivo models like organoids and other types of in vitro. When you build a lot of data sets, you can train machine learning and artificial intelligence models to harmonize the different data sets and try to predict from looking at the preclinical model,

what will happen in the clinical model when you already have a lot of clinical data. And that's kind of the vision for how to grow our artificial intelligence. So you need to be also dedicated not only to profiling human patients, but to also profile lab models. And finally, we acquired a company three years ago in Switzerland called Nebion AG. And Nebion is a company that is 20 years old. It's a company whose headquarters are in Switzerland in Zurich,

operating worldwide and it's a bioinformatics company. So they created capabilities to curate and ingest all the public domain single cell sequencing data. So we took a bet three years ago that we are not going to be the only provider doing single cell sequencing. There are going to be publications in Nature or in Cell or you know, pick your favorite journal. The data is going to be uploaded to AMICA. And so we have a lot of the public domain ingested into AMICA

and informing our decisions and our application of the Immune Dynamics Engine, the idea that I told you about. So those three layers are forming AMICA. Now, immune pathways have a lot of redundancy. So if one pathway is blocked, another pathway takes over. So in your algorithms, how are you dealing with that? that? that? that?

Right, so indeed the question, the fundamental question or one of the fundamental questions is what happens when you block certain pathways or you upregulate or downregulate specific genes and genes are not operating in a vacuum because when you are doing one thing to a specific gene in a specific cell, it's going to have some counter effect or some other effect of the immune system. What we are measuring,

as I mentioned in AMICA, we are not measuring one type of cell or one type of gene. We're measuring in one shot 5 ,000 cells from a human patient sample. And for each of these cells, we are measuring all the genes and all the proteins. And we're going to measure the way that those genes are kind of operating in concert. And we're going to deduce causality rules for how specific genes going up are impacting other genes going up or down. So you are creating the rules of engagement

for how genes and proteins are expressed in immune cells and how immune cells are interacting with one another. And it's really a systems immunology problem and not a cell immunology problem. You're not trying to understand one cell, you're trying to understand the concept of cells and the concept of genes and proteins that are interacting within those cells. And so in terms of like, as you're developing this, right? So what kind of validation have you done to...

to make sure that the insights you've gotten, are reproducible or so?

Right, so I think that it's a very interesting question because you're touching an important point that if your insights are not reproducible, what does it mean? It means that for patients that are taken at random from two cohorts of patients that are given the same drug, you're not going to get the same insights. And if this is going to happen to you, you are in deep shit, pardon my French.

You really want the results, the insights to hold water. You want them to be reproducible. You want the insight that you provide that are going to impact the way pharma companies are treating patients or clinicians are treating patients are going to be robust, reproducible, and you need to validate it really carefully. So when we find insights, first, there are ways, statistical ways to validate this in silico, computationally.

But after you do that, we also validate this in the lab. We have a large lab in New York where our headquarters are. And every insight we're going to come up with, we're going to validate. And it's like really going from correlation to causation. If we think that the modulation of specific gene in a monocyte or a macrophage is critically important for something that should happen in the cell and it doesn't happen in an...

in vitro experiment or in vivo experiment in a humanized mouse, why do you trust this insight? So you need to go back to basic principles. And this is called a lab in the loop. We have a lab that is an instrumental part of who we are as a company. Every insight that we come up computationally, we're trying to validate in vitro, ex vivo, or in vivo, and then bring it back with more data and more insights. So it's a feedback loop. And I think that...

This question that you're asking, I think that most companies are not asking sufficiently well to themselves, do we trust the insights? Is the insight reproducible? And I think it's a critically important question. So thank you for the question. Now, in terms of the business model, Immunai doesn't appear to be building its own pipeline of therapies. So can you explain what the business model is that about service business or how do you maintain a stake in the therapies that you help discover?

Right, so first we view ourselves as design partners, so not as a service provider. We want pharma and biotech companies to come to us with their biggest questions. And maybe I'll say a statement about service providers are usually content by being paid to do a service that is going to solve a very niche application without knowing how this impacts the top line or the bottom line. For us,

we are really trying to create a relationship of trust with our pharma partners to really go all the way to the top and understand what success looks like. So for example, when we are going to optimize a clinical trial that is going to, you know, hopefully lead to a good outcome of an FDA approval and one billion dollar income a year, ideally we would want to get some part of it. As you know really well, pharma companies,

especially the more robust they are, it's very difficult for them to share revenues. So there are different ways to go about it. For example, to talk about milestones, to talk about success -based payments. The earlier you are, if you are providing to them novel targets that are about why certain mechanisms lead to immune resistance, to why you're not going to get clinical response, and you're finding novel targets

that can lead to new drugs and those drugs are being created and you have part of it, then you can really get part of the upside, really part of the drug. So the earlier you are, the better. And we are working on both the discovery and the development, but we have a very strong focus on development and I'll tell you why. I think nine, if not 10 out of 10 platform companies that are using AI for therapeutics,

they are doing it to improve discovery. And the problem that I view in the world of therapeutics today is that you have a lot of novel targets, a lot of novel new drug candidates, but all these companies do not have better statistics in bringing them to FDA approval. And at the end of the day, when I'm giving you a novel target, you have to ask yourself, why do I think this is going to get FDA approval?

So all of our technology and all of our measurement of the human immune system and the human immune response and resistance is geared towards optimizing for FDA approval. How can we get to the outcome of this drug that we are profiling in the lab and in the phase 1 stage is going to lead to an FDA approval? And over time, it's been a few years that we are doing, every year we are growing our customer base. We are growing our top line and bottom line.

And we are getting to the point where they are viewing us really as partners and not just as a service provider. But really in the long run, we are fascinated by being able to create this oracle, this immune intelligence oracle that is able to improve the drug development process and improve patients' outcome. And our business model is about partnering as design partner, as thought partner to the biggest pharma companies, to the biggest biotechnology companies

so that we will have some cut in many drugs. And I'm not trying to optimize to, you know, are we going to own this amount of the drug or that amount of the drug when the brain of this machine is going to be evident, I don't think that this is going to be an issue. So it's like really still building the power of this machine, the immune intelligence being unlocked. So Immunai has raised $215 million in 2021 and...

that's brought the total funding to $295 million. And you've explained what the company is doing. So how have you used this capital to date? Because it's a pretty unique proposition and a pretty unique way in which you're building all of this.

Right, so as I mentioned, we have about 10 ,000 and change samples that we have sequenced and we are aiming to build not one order but two orders of magnitude larger. So it's a very expensive endeavor. And I can tell you that from my beginning as the CEO of Immunai, I was always very concerned about how do we build a company that can last

for the entire endeavor. And this is not going to take another year or two, it's going to take another 10 years. And so I think that the money that we raised in 2021 and before, we are not even in the halfway of using this. So we are very concerned about the cash and are very optimistic that we have a very strong runway to get to the other side, which is, you know, in two years or three years when you interview me again, it may be that I'm going to tell you and now,

the immune intelligence is being proven in ABCD and all of these are very strong case studies of clinical success. And that's where we want to use all the funding for. I think we're doing a very good job in really holding a very tight budget and trying to spend only the necessary amount to keep us going for a few more years. So Dr. Noam Solomon,

co -founder and CEO of Immunai. Thank you for your time.

Well, Amar, what did you think? It was fascinating to hear about Immunai. It's a pretty large undertaking, as Noam described, right? So for each patient, they are profiling thousands of immune cells of different types, and then they're doing this single cell experiments, and then, of course, collecting data

for thousands of genes and proteins. So it is a huge undertaking and it's a pretty vast approach.

Noam talked about the complexity of the immune system. Does complexity alone make this a good area for the application of AI? I mean, so I've worked in immunology and oncology for most of my career. So it's a very complex area. Immunology targets are very hard to find because what you find is you try to modulate in one way, but then...

there are biological pathways that take over on the other side. So there's a lot of redundancy built in the immune system, which is understandable because in the evolutionary approach, you don't want to be one type of germ attacks, you don't want to get killed right away, right? So that's why for, yeah, I mean, it's a challenging approach. I wouldn't say it's necessarily this fits better than in the others. I think like a lot of the areas in, mean, human biology is just complex. I mean, knowledge is very complex, ecology is complex, but a lot of the other areas

are complex as well. So, but I think like these kind of the approaches that Noam talked about, right, those are necessary now to really unravel a lot of these complex issues and really try to find the solutions to the scientific questions that we have.

Well, Noam talked about the questions of how to profile the immune system and the atlas immunize constructing. One consistent theme on the show has been that the quality of the results are going to depend on the quality of the data. What do you think about immunize approach? Yeah, absolutely. I mean, the quality of data is going to be very important here. And especially see yes, I'm there getting

a lot of samples. But on the other hand, see, they also have a lot of features, right? You have like tens of thousands of genes or so, right? These are all the features. And then they talk about measuring the cell surface receptors and then doing the clinical studies, as well as the preclinical studies, right? Animal studies. So you have a lot of features there. So, and to be able to then find the right signals, you're gonna need to have a really high quality data.

Because if you don't have the high quality data and because the number of features are so high, there is a great risk that you're going to get a lot of false positives. So the higher quality the data is, the lower the false positives will be.

You talk about Immunai being an engineering company first. To what extent is this an engineering problem that it's trying to solve? Yeah, it is an engineering problem because Noam talked about this building the systems immunology, right? So not necessarily focusing on a particular gene or protein, but looking at how the entire immune system works. And that's very much

an engineering problem where you have a lot of different components and you're trying to understand how those components are working together so that the system is working fine. And then when you perturb the system, what kind of effects are you going to see and how do you then fix that system, right? So it is a very much of an engineering problem. The problem with biological systems is that there just is like too much variability

in the biological system. So these are very difficult engineering problems, but I'm glad that they are taking this approach with a large amount of data and really tailoring the algorithms for specific problems because I think that's needed.

One of the interesting things you talked about is that Immunai is not only looking at the immune system, but the immune system as it reacts to treatments, which I thought was interesting. But he also talked about the challenge of the large number of immunotherapies given to small cohorts. How does that complicate what they're trying to do? It's again, as I talked before, right? You have a lot of features or like a lot of these different genes. And so these are the features and the number of samples,

Yeah, is limited to the number of patients. So if you have a limited number of patients and you are then providing a lot of different immune therapies, then the number of patients with a specific therapy, that number, that data sample number gets smaller. And that's an issue because what's important is that you need to have a lot of samples. When you have this number of features, you need to have a lot of samples. If you are, you're having different types of samples,

it becomes a harder problem. Single cell technology, he noted, was very noisy. And the fact that RNA degrades rapidly makes this all the more complicated and the need for speed. Leveraging machine learning with deep learning to address those challenges is an approach they're taking. But he also said this requires art, not just science. Is art ultimately going to distinguish AI companies from each other? Is it the art rather than the science that will be the big difference?

It's very important to understand what's actually going on in these cells. And companies that try to understand and have this domain knowledge and incorporate that in their approach, that's the key. You can't just take any new algorithm and just apply it to the biological systems because the biological systems...

I mean, there's a lot that you need to know about those. So, and you know, when I've been doing like analytics, right, yeah, there is art involved in terms of how you actually set up the analysis, what are the type of algorithms that you use. I mean, there is no cookbook for that. You really need to understand what the problem is. You know the knowledge of what type of algorithms there are, what type of problems they solve, and you need to match it with that. So, yes, I mean, you need to set that up. Then in terms of...

actually performing the analytics, yes, that's more of a scientific approach, but how you actually set up the problem and how you set up the solution for that, you do need to be creative about that. So yeah, I agree. I think the way you can be creative and the way you adjust to the problem is going to distinguish the companies that are successful versus not.

Well, it was a great discussion and thanks again. Until next time. Thanks, Dan. Thanks again to our sponsor, Agilisium Labs. Life Sciences DNA is a bi -monthly podcast produced by the Levine Media Group with production support from Fullview Media. Be sure to follow us on your preferred podcast platform. Music for this podcast is provided courtesy of the Jonah Levine Collective.

We'd love to hear from you. Pop us a note at Danny at LevineMediaGroup .com. For Life Sciences DNA and Dr. Amar Drawid, I'm Daniel Levine. Thanks for joining us.

‍

‍

‍