What are the key takeaways from this The Bio Report episode?

Key insights include: **Causal vs. Descriptive Data:** Models trained on descriptive data like single-cell RNA sequencing fail at causal prediction tasks at rates no better than a coin toss. Xara generates genome-scale perturbation datasets using PerturbSeq — 20,000-gene perturbations by 20,000-gene readouts across multiple cell types — to train models capable of predicting which genetic changes transition cells from diseased to healthy states.; **Undruggable Target Strategy:** Rather than competing on accessible targets where existing antibody methods like phage display or humanized mouse immunization already perform well, Xara focuses on multipass membrane proteins, GPCRs, ion channels, and agonist antibodies to heteromeric receptors — categories where conventional methods frequently fail, providing pipeline differentiation and the strongest showcase for AI capability.; **Design-Make-Test Feedback Loop:** Xara's X-Design model evaluates approximately one billion antibody designs computationally, filters candidates through a secondary model, then physically synthesizes and tests roughly one million designs in the wet lab annually. Successes and failures feed back into the model, progressively improving hit quality toward leads and eventually development candidates without manual iteration.

What did A Billion-dollar Bet discuss on The Bio Report?

Marc Tessier-Lavigne, co-founder and CEO of Xara, explains how the company deploys over $1 billion in funding to apply end-to-end AI across three drug development bottlenecks: target identification, molecular design, and patient stratification, with an initial focus on historically undruggable biological targets. Key topics include: **Causal vs. Descriptive Data:** Models trained on descriptive data like single-cell RNA sequencing fail at causal prediction tasks at rates no better than a coin toss. Xara generates genome-scale perturbation datasets using PerturbSeq — 20,000-gene perturbations by 20,000-gene readouts across multiple cell types — to train models capable of predicting which genetic changes transition cells from diseased to healthy states.; **Undruggable Target Strategy:** Rather than competing on accessible targets where existing antibody methods like phage display or humanized mouse immunization already perform well, Xara focuses on multipass membrane proteins, GPCRs, ion channels, and agonist antibodies to heteromeric receptors — categories where conventional methods frequently fail, providing pipeline differentiation and the strongest showcase for AI capability..

How long is this episode of The Bio Report?

This episode is 46 minutes long. SignalCast provides an AI-generated summary so you can get the key insights in about 3 minutes.

The Bio Report

A Billion-Dollar Bet on AI-First Drug Development

January 28, 2026

46 min episode · 2 min read

A Billion-dollar Bet

Episode

46 min

Read time

2 min

Topics

Career Growth, Productivity, Health & Wellness

AI-Generated Summary

Published Mar 14, 2026

Key Takeaways

✓Causal vs. Descriptive Data: Models trained on descriptive data like single-cell RNA sequencing fail at causal prediction tasks at rates no better than a coin toss. Xara generates genome-scale perturbation datasets using PerturbSeq — 20,000-gene perturbations by 20,000-gene readouts across multiple cell types — to train models capable of predicting which genetic changes transition cells from diseased to healthy states.
✓Undruggable Target Strategy: Rather than competing on accessible targets where existing antibody methods like phage display or humanized mouse immunization already perform well, Xara focuses on multipass membrane proteins, GPCRs, ion channels, and agonist antibodies to heteromeric receptors — categories where conventional methods frequently fail, providing pipeline differentiation and the strongest showcase for AI capability.
✓Design-Make-Test Feedback Loop: Xara's X-Design model evaluates approximately one billion antibody designs computationally, filters candidates through a secondary model, then physically synthesizes and tests roughly one million designs in the wet lab annually. Successes and failures feed back into the model, progressively improving hit quality toward leads and eventually development candidates without manual iteration.
✓AI as Biology's Mathematics: Eric Schmidt's framing — that AI is to biology what mathematics is to physics — provides a practical framework for understanding why AI succeeds where equation-based biological modeling failed for decades. Unlike physics, biology has no derivable governing equations, but AI can detect patterns across high-dimensional datasets to build predictive cellular models, making it the correct modeling tool for drug discovery.
✓Bilingual Talent Pipeline: No sufficiently large pool of scientists fluent in both AI and biology currently exists. Xara's near-term solution pairs AI scientists with disease biologists and drug hunters in program-based teams, while building internal AI agents that let each group self-serve in the other's domain. The company anticipates a generation of natively bilingual scientists emerging within ten years from universities and early-career roles.

What It Covers

Marc Tessier-Lavigne, co-founder and CEO of Xara, explains how the company deploys over $1 billion in funding to apply end-to-end AI across three drug development bottlenecks: target identification, molecular design, and patient stratification, with an initial focus on historically undruggable biological targets.

Key Questions Answered

•Causal vs. Descriptive Data: Models trained on descriptive data like single-cell RNA sequencing fail at causal prediction tasks at rates no better than a coin toss. Xara generates genome-scale perturbation datasets using PerturbSeq — 20,000-gene perturbations by 20,000-gene readouts across multiple cell types — to train models capable of predicting which genetic changes transition cells from diseased to healthy states.
•Undruggable Target Strategy: Rather than competing on accessible targets where existing antibody methods like phage display or humanized mouse immunization already perform well, Xara focuses on multipass membrane proteins, GPCRs, ion channels, and agonist antibodies to heteromeric receptors — categories where conventional methods frequently fail, providing pipeline differentiation and the strongest showcase for AI capability.
•Design-Make-Test Feedback Loop: Xara's X-Design model evaluates approximately one billion antibody designs computationally, filters candidates through a secondary model, then physically synthesizes and tests roughly one million designs in the wet lab annually. Successes and failures feed back into the model, progressively improving hit quality toward leads and eventually development candidates without manual iteration.
•AI as Biology's Mathematics: Eric Schmidt's framing — that AI is to biology what mathematics is to physics — provides a practical framework for understanding why AI succeeds where equation-based biological modeling failed for decades. Unlike physics, biology has no derivable governing equations, but AI can detect patterns across high-dimensional datasets to build predictive cellular models, making it the correct modeling tool for drug discovery.
•Bilingual Talent Pipeline: No sufficiently large pool of scientists fluent in both AI and biology currently exists. Xara's near-term solution pairs AI scientists with disease biologists and drug hunters in program-based teams, while building internal AI agents that let each group self-serve in the other's domain. The company anticipates a generation of natively bilingual scientists emerging within ten years from universities and early-career roles.

Notable Moment

Despite twenty years of new modalities — RNA vaccines, antibody-drug conjugates, and others — the core metrics of drug development remain essentially unchanged: roughly thirteen years from target to approval, 90–95% clinical failure rates, and billion-dollar costs per drug, making the status quo economically unsustainable long-term.

Know someone who'd find this useful?

You just read a 3-minute summary of a 43-minute episode.

Get The Bio Report summarized like this every Monday — plus up to 2 more podcasts, free.

Pick Your Podcasts — Free

Keep Reading

Stopping Shape-Shifting Tumors with a First-in-Class Epigenetic Drug

Jun 10 · 28 min

NVIDIA AI Podcast

How AI-Powered Holograms Are Reimagining Fan Experiences at the Big Game - Ep. 288

Feb 4

Rewriting the Rules of Antibody Drug Design

Jun 3 · 37 min

The Long Run with Luke Timmerman

Ep189: Marc Tessier-Lavigne on Reinventing Drug Discovery with AI

Nov 12

Books, tools, and gear mentioned in this episode

SignalCast may earn commission on purchases via these links. As an Amazon Associate, SignalCast earns from qualifying purchases.

Tools

PerturbSeq
“Xara generates genome-scale perturbation datasets using PerturbSeq — 20,000-gene perturbations by 20,000-gene readouts across multiple cell types — to train models capable of predicting which genetic changes transition cells from diseased to healthy states.”
X-Design
“Xara's X-Design model evaluates approximately one billion antibody designs computationally, filters candidates through a secondary model, then physically synthesizes and tests roughly one million designs in the wet lab annually.”

company

Xara
“Marc Tessier-Lavigne, co-founder and CEO of Xara, explains how the company deploys over $1 billion in funding to apply end-to-end AI across three drug development bottlenecks.”