Dec 22, 2022 · 33m · how-i-built-this
HIBT Lab! Immunai: Noam Solomon
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of How I Built This Lab, Guy Raz interviews Immunai CEO and co-founder Noam Solomon about leveraging artificial intelligence, single-cell genomics, and advanced mathematics to map the human immune system and transform drug discovery.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Guy holds 27.9% of the talking time here. How this is scored →
Guy as informed peer 3.5 Guest teaching 3.9 Guest disagreement 0.2 Guy pushing back 0.5
speaking balance: gold is Guy, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 29:56 Dismantling the autonomous vehicle analogy
Noam directly reframes Guy's autonomous vehicle comparison, emphasizing that unlike autonomous driving where human drivers serve as a benchmark, drug discovery has no successful human baseline since 90% of drugs fail.
Hardest push from Guy ▶ 30:58 Guy presses on lingering risk and uncertaintyGuy challenges Noam on why there is still so much talk of high risk and uncertainty given the company is four years in, highly funded, and demonstrating clear progress.
Biggest teaching moment ▶ 10:54 The mouse model fallacy in oncologyNoam educates Guy on the fundamental flaw in modern drug development by quoting a mentor: 'we can cure every type of cancer in mice,' illustrating why animal models fail to translate to human immune responses.
Guy holds their own ▶ 27:25 Guy cites AlphaGo's Move 37Guy displays deep background knowledge of Immunai's published thought leadership, citing co-founder Luis's essay on AlphaGo's Move 37 to query whether AI can discover non-intuitive oncology targets.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Guy as informed peer | Guest teaching | Guest disagreement | Guy pushing back | Why |
|---|---|---|---|---|---|---|
| Origins: From Mathematics to Immunotherapy | 3 | 2 | 0 | 0 | Guy asks open-ended biographical questions about Noam's transition from academic math postdocs at Harvard and MIT to founding a biotech company. Noam explains how his co-founder's grandfather's adverse reaction to cancer immunotherapy inspired him to apply data science to medicine. | |
| Building the 'Google Maps' of Immunology | 3 | 4 | 0 | 0 | Guy asks Noam to explain the 'Google Maps' metaphor for the immune system. Noam educates the host on single-cell technologies that measure 20,000 genes per cell and generate a terabyte of data per patient sample. | |
| Technological Convergence and Personalized Medicine | 3 | 4 | 1 | 0 | Guy hypothesizes that because every human immune system is unique, universal mapping could immediately personalize treatment. Noam gently refines this by comparing immune systems to eye colors—individual but falling into distinct, clusterable categories. | |
| Drug Development Inefficiencies and Animal Models | 4 | 5 | 0 | 1 | Guy asks about immunotherapy development bottlenecks, and Noam outlines the $2.6B cost and 90% failure rate. Noam teaches Guy about the critical translational gap between animal models and human immunology with the insight that science can cure cancer in mice but fails in humans. | |
| Business Model, Revenue, and Industry Collaboration | 4 | 3 | 0 | 1 | Guy probes the business model, noting $300 million raised might not be enough for massive scientific R&D, and asks about competition. Noam explains their fee-for-service revenue model with pharma partners and welcomes industry competition. | |
| Data Ingestion, MICA Database, and Trial Optimization | 3 | 4 | 0 | 0 | Guy asks practical questions about how physical data collection works across global medical partners. Noam details their MICA database covering over 100,000 patients and 500 disease indications across blood and tissue samples. | |
| Responders vs. Non-Responders and Cancer Simulations | 3 | 5 | 0 | 1 | Guy asks if mapping could make all cancers treatable in 50 years. Noam gives an in-depth biological breakdown of how subtle gene variations distinguish drug responders from non-responders and why computer simulations are needed to model cancer escape mechanisms. | |
| Multidisciplinary Teams Bridging Math and Biology | 3 | 3 | 0 | 0 | Guy highlights the synthesis between computational power and biology. Noam describes building cross-disciplinary teams where mathematicians study biology and immunologists learn machine learning. | |
| Embracing Curiosity: 'When Not Knowing Is a Methodology' | 5 | 4 | 0 | 0 | Guy demonstrates preparation by citing Noam's essay on curiosity and co-founder Luis's article about AlphaGo's Move 37. Noam elaborates that immune mapping is far higher-dimensional than Go, requiring AI because human intuition cannot process 20,000 variables across thousands of cells. | |
| Autonomous Vehicles Analogy vs. Drug Discovery | 4 | 5 | 1 | 2 | Guy draws an analogy between data loops in autonomous vehicles and drug discovery, then pushes on why high uncertainty remains after four years of progress. Noam dismantles the AV analogy by pointing out AVs have human driving benchmarks, whereas drug discovery lacks any reliable human baseline. |
Statements from this episode (16)
Solomon: Immunai generates one terabyte of data per patient sample
“We are going to generate one terabyte of information from every sample that we take from a patient.”
Solomon: Immune system mapping enables data-driven drug discovery
“Having a map of the human immune system will enable researchers to apply data mining tools to come up with novel ways to develop and discover therapeutics.”
Solomon: Dramatic compute growth over five years enables immune system mapping
“What happened in the past five years was in some sense a perfect storm where compute power has grown dramatically. We have seen more and more, and we keep seeing more and more technologies in biology and chemistry that allow us to measure things in, you know, …”
Solomon: Immune system mapping will enable treatments for rare diseases
“When we understand the immune system better and better, and we map more areas of the immune system, we will be able to cope with more rare autoimmune indications, orphan disease, and other you know, cancer indications that only few patients have.”
Solomon predicts immune profiling will enable preventative healthcare within 15 years
“The vision that I have and the hope that I have is that maybe in 15 years, We'll be able to use our new profile and capabilities to help people remain healthy.”
Solomon: Immunotherapy development costs $2.6B+ with a 90%+ failure rate
“So to develop a new immunotherapy takes more than 2.6 billion dollars. It takes more than 10 years of research and development before market. And more than 90% of drugs fail To get an FDA approval.”
Solomon: Bridging animal testing to humans is biotech's biggest challenge
“Understanding how to bridge the gap between the human immune response and the way that drugs are being tested in animal models. Is, I believe, the biggest challenge that the biopharma and biotechnology industry should face and will face in the coming, ah, deca…”
Solomon: Immunai generates commercial revenue from large biopharma partnerships
“First is that we are already working and making money by working with our partnerships you know, large Biopharmaceutical companies are paying us good money to help them further develop and improve the development of their drug candidates and other drugs.”
Solomon: Immunai's database spans over 100,000 patients across 500 diseases
“It is the largest of its kind database of the immune system with a single cell resolution. So today we have a database that spans over 100,000 patients, over 500 different disease indications, thousands of studies, and we're trying to put as much clinical cont…”
Solomon: Immunai's platform successfully identifies superior clinical drug combinations
“We were able to show quite definitively that our platform can identify the preferred combination arm that will demonstrate a superior immune response. So it happened already.”
Solomon: Immune Profiling Pinpoints Gene Differences Explaining Cancer Immunotherapy Non-Response
“And the interesting thing is that when we do our immune profiling of patients pre and post therapies, we see that some very subtle nuanced differences between responders and non-responders explain, or at least partially explain the reason. So we can find that …”
Solomon founded Immunai knowing nothing about biology, medicine, or business
“I started the company. I didn't know anything about biology, medicine, business development, and I had to ask a lot of questions”
Solomon: Humans lack the intuition to analyze high-dimensional biological data
“No human being will have the intuition to analyze all the relevant data. You have to use a machine to analyze the data.”
Solomon: Unlike autonomous vehicles, biotech AI must vastly outperform human experts
“For autonomous vehicles, you have experts, and every driver that can drive a car, you can ask whether an autonomous vehicle drives as good as a driver, like me and you. For drug development, even the best scientists in the world, they don't know, as we mention…”