DARPA-Backed ASU AI Maps 300 Million Proteins in an Hour to Outpace Pandemics and Biothreats

DARPA-backed AI targets protein function mapping to bolster biodefense and outbreak response

Speed is the variable that decides outcomes in pandemics and biothreat events. A new Defense Advanced Research Projects Agency (DARPA) effort led by Abhishek Singharoy, associate professor in Arizona State University's School of Molecular Sciences, takes direct aim at that constraint.

His team is building NODES (Network of Optimal Dynamic Energy Signatures), an AI program that predicts protein function by analyzing motion, not just static structure. The goal: evaluate roughly 300 million proteins in about an hour with more than 90% confidence.

Why motion matters

Most models treat a protein like a single still frame. NODES watches the film. By capturing how proteins move and shift conformation, it infers function with higher confidence and can scale beyond the limits of stationary-structure approaches.

Scientists know the sequence and structure of hundreds of millions of proteins, yet functional annotations lag far behind. NODES addresses that gap by marrying physics with AI to read the energetic "signatures" of protein dynamics.

What it could change for research and response

• Threat assessment: Prioritize unknown sequences from environmental or clinical surveillance for lab validation.
• Outbreak triage: Flag candidates involved in host entry, immune evasion, or transmission early in an event.
• Biosafety review: Screen constructs to reduce risks in biological applications.
• Drug discovery: Shortlist actionable targets and mechanism-informed hypotheses faster.
• Disease biology: Clarify roles of orphan proteins implicated in complex conditions.

"If successful, NODES will dramatically improve our ability to assess potential threats from new and unknown protein sequences," Singharoy said. "It will also speed up the process of turning scientific discoveries into real-world applications, help us understand complex diseases, strengthen our defenses against infections and biothreats, and ultimately, protect human lives."

How NODES works at a glance

• Physics-informed AI: Learns energy-flow patterns tied to conformational change.
• Full-motion analysis: Prioritizes dynamic behavior over fixed snapshots.
• Scale and confidence: Targets ~300M proteins/hour with >90% prediction confidence.

Tijana Rajh, professor and director of ASU's School of Molecular Sciences, noted that the project exemplifies interdisciplinary work that pairs deep scientific expertise with modern AI methods to advance human health.

Why this matters now

The speed and scope of NODES are built for critical windows: identifying potential bioweapons before harm spreads, and informing rapid countermeasure development. The same capability can streamline target selection, de-risk programs earlier, and surface new treatment avenues.

Perspective and next steps

As with any predictive system, experimental validation remains essential. The value here is precision triage at unprecedented scale, giving teams a head start when time is scarce.

Learn more about DARPA's mission and programs at DARPA, and explore advances in protein structure resources such as the AlphaFold Protein Structure Database.

Build AI capability in your lab

If you're upskilling teams on applied AI for scientific work, see curated pathways by role at Complete AI Training - Courses by Job.