David Yakobovitch
DUE TO SOME HEADACHES IN THE PAST, PLEASE NOTE LEGAL CONDITIONS:
David Yakobovitch owns the copyright in and to all content in and transcripts of The HumAIn Podcast, with all rights reserved, as well as his right of publicity.
WHAT YOU’RE WELCOME TO DO: You are welcome to share the below transcript (up to 500 words but not more) in media articles (e.g., The New York Times, LA Times, The Guardian), on your personal website, in a non-commercial article or blog post (e.g., Medium), and/or on a personal social media account for non-commercial purposes, provided that you include attribution to “The HumAIn Podcast” and link back to the humainpodcast.com URL. For the sake of clarity, media outlets with advertising models are permitted to use excerpts from the transcript per the above.
WHAT IS NOT ALLOWED: No one is authorized to copy any portion of the podcast content or use David Yakobovitch’s name, image or likeness for any commercial purpose or use, including without limitation inclusion in any books, e-books, book summaries or synopses, or on a commercial website or social media site (e.g., Facebook, Twitter, Instagram, etc.) that offers or promotes your or another’s products or services. For the sake of clarity, media outlets are permitted to use photos of David Yakobovitch from the media room on humainpodcast.com or (obviously) license photos of David Yakobovitch from Getty Images, etc.
Welcome to our newest season of HumAIn Podcast in 2021. HumAIn is your first look at the startups and industry titans that are leading and disrupting ML, AI, data science, developer tools, and technical education. I am your host, David Yakobovitch, and this is HumAIn. If you like this episode, remember to subscribe and leave a review. Now onto our show.
David Yakobovitch
Welcome to the HumAIn Podcast your first look at the newest products in the AI, data science and ML and deep-tech industries with startups. What founders are building to change the world and make it a better place for humans and machines? On today’s episode, we’re featuring Raviv Pryluk, who works at Immunai. He’s focusing on scaling up as an operator, one of the fastest-growing startups in the immunology space looking about how to solve problems with data-driven autoimmunity technology to fight things like cancer and autoimmunity. Raviv thanks so much for joining us on the show.
Raviv Pryluk
Thanks for having me. It’s a great pleasure.
David Yakobovitch
Absolutely. It’s my pleasure as well. I’m a big fan of founders and operators, especially in the silicon nation coming forth and building companies that are between Israel and the United States. So exciting to see everything that you and the company are doing. But first, let’s share with the listeners. Can you tell us a little bit about your story and what you were doing in the technology and then what led you to help scale up Immunai?
Raviv Pryluk
So in my academic background, I did a bachelor’s and master’s in engineering at the Technion Israel Institute of Technology. And in my Ph.D., I switched to computation and neuroscience. So actually, I enjoyed working in the defense industry for a few years. But I was excited about the opportunity to incorporate engineering into biology. And I was always fascinated about what’s going on in our brains. So my Ph.D. is well, it was the first switch or shift from engineering to biology, which was quite a dramatic change from, solving problems, you’re doing one thing and you can expect an outcome.
And then in biology, you can do the same thing over and over again, but the results are changing. So that was in my academic background. And after working many years in the defense industry, someone introduced me to now the CEO of Immunai. And he told me something that convinced me to join the company. It’s great serving the country and working in defense industries, but it’s more important to serve humanity. This is exactly what we are trying to do at Immunai. So that’s me in a nutshell.
David Yakobovitch
Very exciting. And I’m really excited about the entire life sciences industry. In fact, recently at DataPower Ventures, we have made an investment in another life sciences technology company that’s looking at data from research papers to uncover when new drugs and new technologies are coming to market, it actually seems it might be complimentary with the technology that you and your team are working on. So let’s hear more about Immunai then can you tell us first about the company today and what your team is doing to reprogram the human immune system to fight cancer and autoimmunity?
Raviv Pryluk
Sure. So Immunai was founded in 2018, on the premise of combining two authentic pillars, so one is single-cell multi-omics, which is the ability to look in high resolution in cells in the body. And specifically, we’re focusing on the immune system, so cells in the immune system. And the second pillar is, all the buzzwords that we’re hearing about big data and machine learning AI, but we’re truly combining these two pillars.
And the idea is to actually be able to build the largest data set with high resolution about the immune system, and then together with the sophisticated computational techniques to unravel the mysteries behind the immune system in many different indications and scenarios. And the idea is that the immune system is a system and if you really map it and investigate it in many scenarios, you will be able to develop drugs and to improve the health of billions around the world.
David Yakobovitch
This is so important for the research and work that’s being done. Today, I work for a database startup. So I work for one called SingleStore and we focus on in-memory and on-disk data and helping that being real-time. But we’re not specific for any type of data. Now, you’re so focused, in fact, on this use case, that is critical. I mean, it’s life-changing when you can make breakthroughs for these diseases like cancer and neurodegenerative diseases, I think about friends and colleagues in my life that if they had a chance to not have a certain condition, like what would be possible.
So it sounds like there are a lot of opportunities there. And it’s not only business, there’s a lot of technology, there’s a lot of data going into these efforts. Can you speak to us a little bit more about these efforts with your database, your data, and your AI for what’s making this cutting edge research today?
Raviv Pryluk
Sure. So first, you mentioned that the opportunity, so I just want to highlight, 8 of 11 best-selling drugs in the world are immune centric, obviously, many in cancer, but not only now, the top four drugs are immune regulator. The most recent one are COVID-19 vaccines that, shaped the way we are fighting the pandemic. But actually, there are huge challenges in developing drugs to help fight diseases. So first of all, it’s very expensive, and it’s time-consuming, immune-related, we are talking about roughly $3 billion to take back to the market within more than 10 years, and the fellow rates are enormously high, and I really want to change it.
So, the basics, what we are doing is we are building a huge database, we are trying to map the immune system. So we are looking at cells, we have already mapped 10s of millions of cells many from humans, but also from animals. And we are collecting a huge amount of cells and mapping them in multi-omics, which means that we are looking at the RNA expressions, the transcriptome, and the protein level, and many other modalities. And the idea that if you look at many modalities, you will be able to really understand what’s happening in the cell. And you can think of the cell as the basic building block of the immune system. But actually, of other systems, I mentioned the brain. So it’s the same for the brain and other systems in the human body.
So by mapping the cells in different disease indications in different states, and different ages will really try to map the system. While having this huge database, we can start asking fundamental questions. For example, what if we talk about COVID, what causes young people to be able to really overcome COVID so easily, and what causes elders to react so differently, so something about the cells of the immune system behaves differently?
What causes some to respond to immunotherapy, one of the powerful drugs in order to fight cancer, while others are not responding? Unfortunately, a high percentage of patients are not responding. And while using the database, we can ask those fundamental questions, and then be able to discover new targets.
So cells are composed of many genes. So we can think about the gene as a target. And then you can surface new targets that you want to target or to attract in order to fight different diseases. So one way to do so is, using very simple computational methods, for example, just using linear algebra, or looking at whether something is high or low, and comparing different expressions, but you can use this sophisticated computational approach that is available today.
And when you have such a huge database, it’s better to do so. So we are applying sophisticated methods in order to query this database. But actually, even while building the database, for example, while collecting the data, some of the data is coming from blood samples, some from tissues, and other sources. We are really trying to make sure that the data is clean. And even in the pipeline of producing the data, we’re using various sophisticated computational methods.
David Yakobovitch
What I find so fascinating is to think back to where technology research was only a few years ago where it was all very wet lab-focused AI-powered. And today it’s become very data and AI-powered, where you can still do a lot of research there. Raviv you mentioned, your team’s doing a lot of simulation, and a lot of synthetic research with these techniques around data, ML, and AI, which helps accelerate the research to create these new breakthrough solutions.
Raviv Pryluk
That’s correct. So what’s exciting about Immunai, is actually, with the biology components live, great scientists that are working in the lab, the best biologists in the world, they’re conducting experiments, and experiments are very complex and take a lot of time. And if you really want to map a system, like the immune system, which is anonymous, or highly complex, let’s say, we are debating a team, whether it’s infinite, complex, or just very complex. But if you really want to map the system, you need to conduct a lot of experiments, to run many, many samples, in many conditions, with different assays. And this is very time-consuming, and also expensive.
But imagine that you can learn from every experiment that you’re conducting, what went right and what went wrong, depending on the question at hand. For example, imagine that you are trying to mimic a specific component of the immune system under a specific condition. And you saw that the results are not exactly as expected. And then you can use simulation in order to simulate what you should do in the next experiment. And then you can fine-tune the experiment and to save a lot of resources, and to be able to reach the targeted point, much faster and accurately. So definitely, combining biology with software engineers, machine learning engineers, and many, many others.
David Yakobovitch
And similar to other industries, where we’re seeing a lot of this type of work, I would even call it digital twins, where we’re seeing that with the manufacturing and the property technology industries. This kind of simulation is also maybe even digital twins, would you say for immunology to see what if there was a David, and he had a condition, and what would happen with certain treatment and go through the cause, but we’re gonna, we’re gonna run this simulation on David 10,000, a hundred thousand, a million times, and go through all these different pathways to see if we can come up with a better solution.
Raviv Pryluk
Totally, and in machine learning, transfer learning is a very powerful tool, under the image, shield, being able to, to use knowledge from one domain, or one context. So, using the pixels from a specific image in order to transfer to another image, helps a lot in this field, and it’s highly applicable and relevant to immunology. So if you think about it, the immune system is composed of cells, each cell is composed of many genes, but the same cells are participating in different conditions.
You mentioned David, so, different David has similar cells, but when you can simulate David under a specific condition, and then transfer it to a different condition, or to a different patient, it is very powerful. And actually, we have a very promising result we are predicting, and then we can test. And I assume we will talk later about testing, and we can test correlation. But what is highly exciting is that we can also test for causation, we can predict something, and then we can actually test whether it’s working or not.
David Yakobovitch
And it sounds like you want to get to this place of correlation. And causation. First requires a deeper understanding, as you mentioned, Raviv on these cells, and what does this look like creating like a dictionary, a taxonomy, or a library of cell types?
Raviv Pryluk
First of all, I told you I switched to biology, so I didn’t know anything about biology. And still, I wouldn’t consider myself as an expert at all. But Tim and I, we have great immunologists starting from our CSO, always an icon. He really worked with cells for many, many decades. And the cell is being represented by the expression of roughly 20,000 different genes, many other components, but mainly the genes, and then different gene expressions represent different cell types. So, back then, people were looking at the expression and would say, based on a specific market, let’s say a specific gene.
If this gene is very high, it means that this is probably a specific cell type, let’s say, we have T cells, so CDA Temora cells, this is the name of the cell in the immune system. And there were different markers. But what we did was not only ask the field by using machine learning, so you can use different cluster algorithms. And then using neural networks or other approaches, you can really learn for many, many repetitions of gene expressions and cell types and really learn the phenotypes of different cells.
And today, we can, through our pipeline, get a new data set and fairly quickly know exactly which cell which is something very powerful because again, this is the building block of the system. And if you really understand that, the cell resolution, what’s going on, you can afterward predict the phenotype and the response of the system as a whole.
David Yakobovitch
And specifically, what you’re hinting at Raviv, is about this prediction, this response, this machine learning that we can look at, based on understanding about this segmentation and metadata about the cells. Well, this leads to the question earlier: What is correlation and causation? Can we actually take that and apply that to the immune system?
Raviv Pryluk
Thanks for the question. So most of the work is observational or collaborative work, which is obviously very important. So let’s take an example, we can look at a data set where we have different patients that go to specific treatment, let’s say immunotherapy. And then we are analyzing, for example, PBMCs, the peripheral blood system, the peripheral immune system. And we can also analyze, for example, tissues to tissues. And we can look before the treatment and different time points along the treatment. And then obviously, some of the patients are responding and others not responding.
And then we can look at the expression of different cells along the different trajectory. And what we can mount is that different phenotypes between those that respond and those that do not respond, but is observational data, it means that, there is something correlative between the responders versus non responders in a specific phenotype. And then you can have disease, for example, you can say, the simple case would be that PD one a very well known healthcare, is related to response. And then you can say, okay, let’s knock out on simple words, for example, let’s shut down this gene.
So the prediction would be that shutting down or turning on a specific engine would change the behavior of the system. Up to this point, you have some hypotheses. But what we can do today with very sophisticated tools, is to actually test this in the lab. We are using CRISPR screening or CRISPR technology in order to knock down a specific gene and then you can see how the system is changing. And we can do anything in a high throughput manner.
So we can shut down many, many genes, one after the other in different environments in different conditions. And then you can really test your hypotheses. And here we are already talking about possession because we did something and we can really see what he’s doing to the system. And we are trying to close the loop from observational data, clinically relevant thought experiments that are more controlling your lab and then going back to the clinics, after we identify the hypotheses, test it, validate and then we can start developing a specific cure.
David Yakobovitch
Something else that catches my mind, having worked on a variety of data analytics and data science projects, is that standards often don’t exist. When we’re thinking about labeling, when we’re thinking about protecting data, when we’re thinking about building responsible data systems by design standards have to be created for a variety of stakeholders. What is your team’s Amina doing around standardization? And how is that going? Because it’s also in the health industry, where there are a lot of barriers to creating standards.
Raviv Pryluk
And this is a very important question. So, we are having many internal discussions around how to improve the robustness of the data at the first level but obviously the robustness of the findings. And we created many different tools and methods in order to make sure first of all the data is as clean as possible. From the way we are running the experiments in the lab, but up to all the upstream analysis, we are conducting a very use of approaches in order to test for statistical significance.
And we are using a lot of different approaches in order to, to make sure that our findings are not biased. And by now, I can say that, compared to all other methods that are out there we are really outperform, and this is something that is crucial for us, because we really want to make sure that when we are making a prediction, it is biologically meaningful because I totally agree with you, I’ve worked in many different fields. And, garbage in garbage out is a very dangerous phenomenon.
But when we are talking about biology, it’s even more because sometimes the time that passes from prediction until you can really test it is very long. And you can spend a lot of resources, both internal people that are walking, but also money, until you find out that the hypothesis was wrong. So we’re really creating a lot of testing and internal tools in order to make sure that finding is robust.
David Yakobovitch
So a big question I have, when I’m looking today, at the evolution of the data, the ML, the AI ecosystem, there’s a lot of talks about AI ops, and ML ops and Data Prep ops, and we’re seeing a lot of tools coming out there. And often companies are coming in one of two directions to attack the problem. Direction number one is, let’s go open source, let’s use this tool, or let’s buy a tool, maybe there’s a tool on the market that already helps with what you’re trying to accomplish.
And the second direction, which I’ve heard from more companies, recently than I expected, is, we’re gonna build the tool, this tool doesn’t exist for our industry, or our use case, we’re gonna build it from scratch. And it sounds like you might be doing a little bit of both with your team Raviv. So I’d love to hear about your thoughts on using open source tech or building and buying tools as well.
Raviv Pryluk
So it’s true, we are doing both the way that we are thinking about this is first try what is out there, we really don’t want to invent the wheel, when it is needed. And especially when we’re thinking about your previous question, if you really want to have robust findings, it’s better using tools that were already validated and being used by many, many others in many fields. It saves you a lot of time of validation, because developing a new tool, and to be really confident that the tool does what you’re expected to do. Now, you need to give it some time on the road.
But obviously, after we’re trying, I don’t want to say all the tools that are available, but most of the tools that are relevant to the problem at hand. And if we eventually find out that it’s not working, or it’s not what we need, we are developing our own tools. But I would say that the Immunai team is very multidisciplinary. So people are coming from many fields, engineering, computer science, math, fields.
So it means that when we are trying to tackle a problem, you have many different directions that people can bring. And usually, something already exists, which obviously needs modification and fix to the problem to the specific data set or the specific features that we have. But after some modification, usually it works. But yes, there are specific cases that we are developing on tools.
David Yakobovitch
You’re taking the right mindset hereof, well, hey, if it exists, like let’s try it out. Let’s see if it works. And if it does work, we’ll use the tool. If not, we’re going to build it. What’s your take there? Get some more comments.
Raviv Pryluk
Yes, so I did do say that what we are trying to do is already so complex, so whenever we can, we are really trying to make our life easier, because at the end of the day, what we are trying to do is not to develop cool tools and techniques, we are doing so but this is a bonus. What we are really trying to do is to cure patients. And, our approach is if we can do it with stones or with linear algebra, let’s do it, let’s kill patients around the world. That’s the mission and the tools are not the mission. And we are really trying to educate the people at Immunai that anytime I come from academia and I publish papers, I want to use cool tools but only if necessary. So this is a very important mindset for us. We are looking for people that are trying to cure patients.
David Yakobovitch
I love it. When you think about it, it’s, it’s great if it’s cool, but if it doesn’t actually do anything it doesn’t matter. So the end goal, the mission to cure patients, get them to a better health status. And that can sometimes mean practical. And like you said, maybe we don’t need all just make it up neural networks, maybe linear algebra works.
So it depends on those use cases. And, there’s a lot to be said there, because we’ve seen in a lot of industries today, there’s, you throw more data at the problem to try to solve it, or you throw more algorithms or better compute or stronger algorithms at the problem to try to solve it. And this brings up the question as well, for the immunology industry, is the solution, like let’s multitask and transfer learning models, which your team is working on? Is it just the data, let’s just get massive data capture? So we can understand more? Or is it a combination of both?
Raviv Pryluk
So this is a great question, we’re debating all the time what is more important, the data or the algorithms, and I will tell you something, an algorithm can extract as much as exists in the data, you cannot extract more than what the data is, is hitting or hiding. And therefore, the data is, to me at least, more important. I want to say that algorithms are not important. Definitely, I have concrete examples of cases where only very sophisticated algorithms allowed us to unravel things that others couldn’t.
And this is actually, very exciting that you are looking at the same data set, but really be able to unravel a specific phenotype or mechanism that you couldn’t without a very sophisticated algorithm. But in general, we are trying to build the richest and the most relevant data set because we believe that the data will allow us to address the questions that we need.
David Yakobovitch
I undoubtedly completely agree there. It’s definitely the combination of the two. And it’s not only technology, we spend a lot of time on the show and our episodes talking about technology. We talk about everything from the data and algorithms.
And we talk about the business. But it goes beyond that. It goes about building and scaling a team. And I know Raviv of your team and Immunai are growing. And you’re building and leading multidisciplinary teams, can you tell us more about your culture and the teams that you’re building and managing?
Raviv Pryluk
So first of all, we are really trying to bring the best people in the world, people that are team above and really trying to play as a team, because none of the problems that we’re trying to solve can be solved alone. And, a year ago, we did a reorganization and we changed the way we are structured in the sense that we moved from being operating on in a disciplinary manner.
So, we have computational biologists, immunologists, software engineers, machine learning engineers, data scientists, and so on and so forth. So you can build teams around the disciplines. But if you are really trying to solve complex problems, as we are trying, the way to solve it is by building groups. So, inspired by Spotify, Wix and others, we have groups in guilds. So our groups are composed from many disciplines.
So for example, the group that I’m leading, you have functional genomics scientists, you have machine learning engineers, computational biology, software engineers, and more. And all of us are fitting together and looking at the problem from different angles. Each one of the team members is bringing his own perspective tools. And at the end of the day, that’s the way to solve complex problems otherwise, in how we couldn’t make the progress that we are making. And that’s the way that we’re trying to build the company.
David Yakobovitch
And so let’s talk a little bit about growth at Immunai, last year, your company raised a series B of funding you’re a venture back startup you’re growing, can you tell us a little bit more about that growth, those opportunities and, and how you’re seeing that growth, transform the company and the industry?
Raviv Pryluk
So, very humbled but what we’re very proud of being able to grow because we were able to convince ourselves that we’re taking the right approach towards changing the way drugs are being developed. And we really believe that the combination between those two pillars of single cell multi omics and, and machine learning AI could revolutionize the way that drugs are being developed. And we are now growing.
There are growing pains. It’s hard but we are now overcoming the challenges and Really trying to grow the teams and the capacity of the things that we’re doing. We’re obviously always hiring talented people, scientists, operators, across many fields. And we really want to be, 10 years from now, the largest pharma tech in the world in order to cure as many patients as possible.
David Yakobovitch
That’s a bold and proud vision in the next decade to be the largest pharma company in the world to cure patients and then help them through their conditions and diseases. That’s very exciting and very noble Raviv. Can you share more with our listeners, what are some calls to action and next steps you’d like to share with the audience?
Raviv Pryluk
So we are now building our platform, which means that we are adding more cells to our database while building our technology capabilities. So we are adding more cool techniques, both in the lab, but also computationally. But more importantly, we are now using the platform in order to develop drugs.
We have some candidates already. We really hope that in a short time frame from now, it starts proving and not only in silico and in vitro and in vivo, as we have shown by now that the platform is working, but also in the clinics, in different phases. Pre-approval and, that’s the main, next steps we really want, to showcase that the platform allows us to develop a better drugs
David Yakobovitch
Excellent. Well, this has been very insightful, not just for myself, but also for each and every listener. So we give thanks to you. The vice president of operations at Immunai. Thank you for joining us on HumAIn. We are very excited to see where the technology transforms, not only this year, but over the next decade.
Raviv Pryluk
David I really enjoyed talking to you, thank you.
David Yakobovitch
Thank you for listening to this episode of the HumAIn Podcast. Did the episode measure up to your thoughts and ML and AI data science, developer tools, and technical education? Share your thoughts with me at humainpodcast.com/contact. Remember to share this episode with a friend, subscribe, and leave a review, and listen for more episodes of HumAIn.