Federated Learning Gains Momentum in Drug Discovery as Pharma Giants and Biotechs Build Collaborative AI Networks
核心洞察
Federated learning enables collaborative AI model training across organizations while keeping sensitive data secure within each institution, addressing IP and privacy concerns.
Eli Lilly's TuneLab initiative uses federated learning to allow small biotechs to access and improve the same models Lilly uses internally.
The AI Structural Biology Network now brings together nine top pharmaceutical companies, demonstrating federated learning at an unprecedented scale.
Artificial intelligence has become a cornerstone of modern drug discovery, but as models grow more sophisticated, the quality and diversity of training data has emerged as an equally critical competitive advantage. The fundamental challenge is well known: pharmaceutical companies, biotechs, and research organizations hold vast amounts of valuable data, yet concerns around intellectual property, patient privacy, and commercial sensitivity make sharing that data difficult. Federated learning is now emerging as a powerful solution to this impasse.
Rather than moving sensitive datasets between organizations, federated learning enables AI models to be trained collaboratively while the underlying data remains securely within each organization. The result is the opportunity to build more accurate, more robust AI models without compromising data ownership or confidentiality.
A shifting landscape for collaboration
While the concept is attracting significant attention, understanding of what federated learning is—and who can benefit from it—varies widely across the life sciences industry. Eight experts joined Scientific Computing World at an exclusive roundtable to discuss their perspective, based on years of operating experience.
Niña Cortina, Co-Founder of LiVeritas Biosciences (搜索), explained that her organization is focused on data generation layer data analysis. "We're a company that can produce the type of data that we could contribute to federated learning, but we're also the builder of computational tools that can either consume or contribute to federated models," she said. Her perspective highlights an important shift: federated learning is no longer simply about pharmaceutical companies sharing information; it is creating opportunities for specialist technology providers, contract research organizations, and biotechs to participate in collaborative AI ecosystems.
Robin Roehm, CEO and Co-Founder of Apheris (搜索), described how federated data networks are enabling organizations to work together on some of the industry's most challenging problems. From the AI Structural Biology Network, which brings together nine of the world's top pharmaceutical companies, to initiatives exploring antibody developability and ADMET prediction, federated learning is beginning to support collaboration at an unprecedented scale.
Large pharma extends capabilities to smaller innovators
Large pharmaceutical companies are also offering these capabilities to smaller innovators. Jonathan Gilbert, Senior Director of Ecosystem Growth and Contributor Partnerships at Eli Lilly and Company, is leading the growth of TuneLab, an initiative built on federated learning. "We're using federated learning to engage with small biotechs for them to use the same models that Lilly uses internally," he said. "They can also improve those models by contributing data in a privacy-preserving way."
For many smaller companies, this represents a significant change in perception. Federated learning has often been viewed as something reserved for organizations with enormous computational resources and extensive internal datasets. Increasingly, however, the technology is being designed to lower barriers to participation, allowing organizations with specialized expertise or unique datasets to make meaningful contributions.
Building connected ecosystems
Technology providers are focused on turning promising concepts into practical research tools. David Gosalvez, Chief Strategy Officer at Revvity Signals (搜索), traced the origins of today's momentum back several years, when Lilly began exploring ways to securely share AI models with collaborators. "There's a real momentum behind many-to-many federated learning networks—not just large pharma, but potentially hundreds of companies contributing, even small ones with unique data contributions," he said.
It is a vision that could fundamentally reshape how drug discovery AI is developed. Instead of isolated organizations building models independently, federated learning offers the possibility of connected ecosystems where knowledge can be shared, improved, and expanded while protecting each participant's most valuable asset: their data.
Key questions ahead
However, important questions remain: how can organizations establish trust between collaborators? What governance models are required? Which technical standards are emerging? And what practical steps can smaller organizations take to become part of federated learning networks?
These questions are explored in depth in The Path to AI Federated Learning for Drug Discovery, an exclusive Scientific Computing World white paper produced in partnership with Revvity Signals (搜索), featuring insights from leaders at companies such as LiVeritas Biosciences (搜索), Apheris (搜索), Eli Lilly and Company, and AstraZeneca.