298: 10,000 Structures Are Not Enough: The Data Gap Limiting De Novo Antibody Design with David Younger - Part 2

The entire public record of antibody affinity data comes to about half a million measurements, and by David Younger's estimate, upwards of 75% of it is AlphaSeq data from his own company's publications. His conclusion is that no single company can close the gap alone. So he is asking pharma and AI labs to pay for an experiment and share every result with the other members.
In Part 2 of this two-part conversation, David Brühlmann talks with Younger, founding CEO of A-Alpha Bio, whose AlphaSeq platform uses engineered yeast to measure protein interactions at scale. Part 1 set out the shortfall: about 10,000 antibody-antigen structures in the Protein Data Bank against a need of hundreds of thousands to millions. Here Younger explains how the consortium keeps proprietary targets out of the shared pool, who has joined, and why he believes it differs from federated learning.
Topics discussed:
• Challenges faced by individual companies in generating sufficient data for effective AI models (02:17)
• How a data consortium enables pre-competitive data sharing while protecting proprietary information (06:13)
• Profiles of ideal consortium members and examples of current participants from pharma and AI model builders (08:36)
• The consortium’s current focus on VHH antibodies and plans for expansion to broader datasets and applications by 2030 (10:00)
• Comparison between the consortium approach and federated learning models in data sharing and collaboration (12:01)
• The main takeaways about the quantity and quality of data needed for machine learning in protein engineering (14:09)
Smart insight: As David Younger envisions it, successful maturation of these consortia could fundamentally shift the field by 2030. Multiple consortia could address different scaffolds, data types, and problem statements, alleviating the bottleneck that has long constrained progress. The hope is that abundant, high-quality data will enable breakthrough therapeutic advancements while revealing new gaps that, in turn, inspire the next wave of innovation. Notably, David Younger sees collaboration—not just data scale—as the true secret weapon. Unlike federated learning, consortia incentivize members to share not just data, but experimental insights and best practices, effectively lifting all boats and accelerating collective progress.
Curious how others are tackling the data problem in protein design? These episodes pick up right where David Younger leaves off.
• Episodes 185 - 186: How Generative AI Accelerates Protein Design for Biotech Discovery with Elise de Reus
• Episodes 115 - 116: Revolutionizing Biologics Development with Hyper Throughput Screening and AI with Jeremy Agresti
• Episodes 123 - 124: Manufacturability: Why Most Protein Candidates Fail (And How to Pick Winners Early) with Susan Sharfstein
• Episodes 219 - 220: From 10,000 Structures to 1.8 Billion Interactions: Breaking the Data Bottleneck to Engineer Efficacious Therapeutics with Troy Lionberger
Connect with David Younger
Company website: www.aalphabio.com (aalphabio.com)
LinkedIn: www.linkedin.com/in/david-a-younger/ (linkedin.com/in/david-a-younger)
Data Atlas website: www.atlas.aalphabio.com (atlas.aalphabio.com)
Support the show (buzzsprout.com/2263581/support)