Autonomous Agentic Workflows for High-Throughput Protein Design and Validation

ABSTRACT

Recent advances in de novo protein design have vastly outpaced manual experimental validation, creating a severe bottleneck in the engineering cycle. We present a fully autonomous, multi-agent workflow that bridges computational design with robotic cloud-lab execution. A central 'Orchestrator Agent' generates protein sequences, communicates via API with automated DNA synthesis providers, and directs robotic liquid handlers to express and purify the designs. Our end-to-end agentic protocol reduces the design-to-validation cycle from weeks to 48 hours, enabling the continuous, autonomous optimization of binding affinities without human intervention.