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

Episode Summary

Executive Summary: Marc Tessier-Lavigne traces his path from a military-family upbringing and philosophy-infused neuroscience training to biotech leadership at Genentech, Stanford, and now Xaira Therapeutics. He explains Xaira’s ambition to use AI plus large-scale causal biology data to design better protein drugs, prioritize targets, and ultimately make drug R&D faster, cheaper, and more predictive.

Main Topics: Early life, schooling, and scientific formation (Priority: 4/5): Tessier-Lavigne describes growing up in a military family in Canada and Europe, attending French schools, and developing an early love of math, physics, philosophy, and later neuroscience. From physics to neuroscience (Priority: 5/5): He explains why he moved from physics to physiology/philosophy at Oxford, how that led him into neuroscience, and how he became fascinated by brain wiring and development. Academic neuroscience and the turn toward translation (Priority: 5/5): His work on axon guidance, retina signaling, and brain development led to the discovery of brain wiring factors and showed him that basic science could have direct human impact, especially for spinal cord injury and stroke. Biotech leadership at Genentech and beyond (Priority: 4/5): He recounts moving to Genentech in 2003, learning drug discovery and executive leadership, then returning to academia while staying involved through boards and co-founding companies like Denali. Stanford controversy and transition (Priority: 3/5): He addresses the Stanford investigation into a few papers, saying his scientific integrity was validated and emphasizing that the episode led him toward a new opportunity at Xaira. Xaira Therapeutics’ AI-driven platform (Priority: 5/5): He outlines Xaira’s strategy: use generative AI for protein design and causal biological models for target discovery, combining platform and product goals with major capital and a large interdisciplinary team. The future of drug development (Priority: 5/5): He argues AI should help cut drug-development timelines in half and raise clinical success rates substantially, enabling a broader era of more effective medicines.

Key Arguments: Problem-solving disciplines such as math, physics, and analytical philosophy shaped his scientific mindset more than rote learning did. Oxford was the pivotal environment where physiology, neuroscience, and philosophy converged into a vocation in neuroscience. Basic neuroscience can be deeply translational; discovering guidance cues and wiring mechanisms can eventually help patients with paralysis or stroke. Industry appealed because it offered a better chance to test whether scientific ideas could become therapies and because progress is more measurable than in very open-ended academic settings. Xaira’s strategy is not just to optimize one step, but to address target selection, molecule design, and patient matching together. Generative AI can move drug discovery from screening “needles in a haystack” to designing the molecule directly. The company’s AI models depend on high-quality public protein structure data plus massive wet-lab design-make-test cycles. For target discovery, descriptive datasets are not enough; causal perturbation data are needed to tell drivers from passengers and predict how to restore diseased cells to healthy states. Xaira is intentionally both a platform company and a product company so that model improvement is grounded in real therapeutic programs. The industry should aim to cut average time from target to FDA approval from about 13 years to roughly 6.5 years and raise clinical success from about 10% toward 20-40% over the next decade.

Data Points: Committed capital: $1 billion - Xaira Therapeutics debuted with committed venture capital in April 2024. Timeline from target to FDA approval: about 13 years - Tessier-Lavigne cites this as the current average drug-development timeline. Clinical success rate: about 10% - He says roughly 1 in 10 drug candidates succeed in the clinic today. Desired timeline reduction: cut in half to 6.5 years - His aspirational goal for AI-enabled drug development over the next decade. Desired clinical success rate: 20% to 40% - He argues the industry should aim to double or triple current success rates. Employee count: 148 - Approximate size of Xaira at the time of the interview. Sites: 3 - South San Francisco, Seattle, and London. Public interest in perturb-seq data: 64,000 downloads in three months - Response to Xaira’s publicly released data sets. Largest perturb-seq datasets published: June publication; largest to date - Xaira released methods and two cell-line datasets to support causal biology modeling. Scale of cell readout in perturb-seq: 20,000 genes - He describes perturbing genes individually and reading out effects across the genome. Scale of design generation: millions of designs - Xaira is generating antibody/protein designs at very large scale. Scale of wet-lab testing: hundreds of thousands tested - He says many designs are tested in the lab and fed back into models. Oxford degree program: PPP (Philosophy, Psychology, and Physiology) - The program that shifted him into neuroscience. Genentech start year: 2003 - He joined Genentech then and later oversaw a large research portfolio.

Pivotal Quotes: "With generative AI, we're trying to design the needle." — Marc Tessier-Lavigne: Explaining how Xaira uses AI to create antibodies and proteins rather than only screen existing ones. "The model can't predict because the model doesn't know what it doesn't know. It's never seen causal data." — Marc Tessier-Lavigne: Justifying Xaira’s focus on large causal perturbation datasets for target discovery and biology modeling. "We should aspire to cut the timeline in half. We have to go from target to FDA-approved drug in 6 and a half years." — Marc Tessier-Lavigne: His vision for how AI should transform industry-wide drug development efficiency.

Implications: Xaira aims to redefine biotech by pairing AI with causal biology and real drug programs. If successful, listeners should expect faster target validation, better-designed therapeutics, and a higher-probability, lower-cost path to new medicines.