E1089 The Next Unicorns E11: insitro CEO Daphne Koller is revolutionizing drug discovery via machine learning & data, shares insights on remote education from time at Coursera & more

Episode Summary

Executive Summary: Daphne Koller discusses the pandemic’s forced reset of education and her current company, Incitro, which applies machine learning and new biological data-generation methods to drug discovery. She argues that biology has lacked large, useful datasets, and that combining wet-lab innovation with AI could reduce drug-development failures, improve precision medicine, and accelerate treatments for major diseases.

Main Topics: Coursera and the limits of online education (Priority: 5/5): Koller reflects on Coursera’s success with adult learners but notes it did not fundamentally change traditional university teaching. She argues MOOCs were ahead of their time and that universities need blended, community-driven, synchronous online learning rather than passive lecture videos. Pandemic-driven transformation of higher education (Priority: 5/5): COVID-19 forced universities to teach online, exposing the weakness of trying to recreate large lecture halls on video. Koller predicts a permanent shift toward more interactive and hybrid models, especially for graduate and continuing education. Incitro’s mission: machine learning for biology (Priority: 5/5): Koller explains that Incitro is building both the data and the tools needed to modernize drug discovery. The company aims to create high-fidelity biological datasets and apply ML to predict disease pathways and therapeutic response earlier in development. Why drug discovery is so expensive and failure-prone (Priority: 5/5): She argues the core problem is not regulation but failure rates: most drug programs fail late, and those failures are extremely costly. Better predictive modeling could help stop dead-end programs earlier and lower the effective cost of successful drugs. Target areas: NASH and central nervous system diseases (Priority: 4/5): Incitro’s first major disease efforts include NASH with Gilead and CNS disorders such as autism, depression, Parkinson’s, and neurodegeneration. Koller emphasizes using patient-derived cells and organoids to identify disease phenotypes and screen for reversals. Ethics and tradeoffs in human trials (Priority: 4/5): Koller discusses vaccine challenge trials and the tension between individual risk and societal benefit. She argues informed adults should be allowed to volunteer for high-risk studies, while acknowledging medicine’s strong no-harm ethos. Future of longevity and precision medicine (Priority: 4/5): Koller does not expect a single cure for diseases like cancer or Alzheimer’s; instead, she expects steady progress through disease subtyping and targeted therapies. She is optimistic that people will live healthier, longer lives, even if radical lifespan extension remains unproven.

Key Arguments: Coursera succeeded primarily for self-directed adult learners, not for transforming traditional university education. Online learning works better when paired with community, deadlines, and teacher interaction rather than fully self-paced isolation. The pandemic may permanently improve higher education by forcing institutions to adopt better digital and hybrid teaching models. Biology needs large-scale, high-quality data before machine learning can meaningfully advance drug discovery. Drug development is expensive because failure rates are extremely high; about 95% of programs fail. Incitro is not a tools vendor but a drug discovery company building a new, data-centric operating model. Smaller, well-defined patient populations can be better targets for drugs than broad blockbuster indications. Big pharma struggles with machine learning because of siloed data, old infrastructure, and weak talent incentives. Organoids and patient-derived cells can reveal disease phenotypes that animal models often cannot capture. Medical ethics should carefully weigh risk and benefit, and informed volunteers should be able to choose to participate in challenge trials. Cancer and Alzheimer’s will likely be addressed through many targeted advances rather than one breakthrough. Interdisciplinary talent that combines biology, data science, and humility is crucial for Incitro’s success.

Data Points: Coursera launch timing: About 7 years earlier - Koller recalls first meeting the host around seven years prior, when she was working on Coursera. Incitro age: About 2 years old - She says the company recently had its birthday and has spent the first two years building infrastructure. Drug program failure rate: ~95% failure rate - Koller uses this to explain why drug development is so costly and slow. LinkedIn members: More than 690 million - Mentioned in the sponsor read for LinkedIn Jobs. Zendesk startup offer: 6 free months - Zendesk for Startups program offer described in the sponsor segment. Vanta discount: $1,000 off - Sponsor offer for Twist listeners. LinkedIn job post discount: $50 off first job post - Sponsor offer for LinkedIn Jobs. NASH full name: Non-Alcoholic Steatohepatitis - Defined as liver inflammation and fibrosis tied to obesity and insulin resistance. Autism twin concordance: 60–70% - Koller cites this as evidence of a strong genetic component. Depression twin concordance: About 50% - Used to support the genetic contribution to neuropsychiatric disorders. COVID vaccine challenge-trial participants: 30,000 people - The host references a vaccine trial example during the ethics discussion. COVID challenge-trial compensation: $2,000 - Host cites a participant payment example in the ethics debate. 20X programs: 20 failed programs - Koller describes the cost burden successful programs carry from many failures. Cystic fibrosis impact: ~90% of patients - She says current targeted treatments have effectively normalized life for most CF patients. Average SOC 2 readiness with Vanta: 2 to 4 weeks - Sponsor read compares Vanta’s speed with a 3 to 5 month manual process. Manual SOC 2 timeline: 3 to 5 months - Used in the Vanta sponsor segment.

Pivotal Quotes: "What you really need is to create some kind of community with a synchronous experience that keeps people moving along at a certain pace." — Daphne Koller: Explaining why MOOCs alone did not solve motivation and completion problems. "We are making data. We are also building tools, both tools to create data and tools to interpret data." — Daphne Koller: Describing Incitro’s core strategy for modernizing drug discovery. "The fundamental problem is that most drugs fail." — Daphne Koller: Explaining why drug development is so expensive and why predictive modeling matters.

Implications: The conversation points to a future where biology is treated like a data problem: better datasets, AI, organoids, and patient-specific models could lower costs and improve precision medicine. Education may likewise shift toward hybrid, community-based online learning.