Accelerating primer design for amplicon sequencing using large language model-powered agents

The pre-trained knowledge compressed in large language models is addressing diverse scientific challenges and catalysing the progression of autonomous laboratory systems, synergized with liquid handling robots. Here we introduce PrimeGen, an orchestrated multi-agent system powered by large language models, designed to streamline labour-intensive primer design tasks for targeted next-generation sequencing. PrimeGen uses GPT-4o as a central controller to engage with experimentalists for task planning and decomposition, coordinating various specialized agents to execute distinct subtasks. These include an interactive search agent for retrieving gene targets from databases, a primer agent for designing primer sequences across multiple scenarios, a protocol agent for generating executable robot scripts through retrieval-augmented generation and prompt engineering, and an experiment agent equipped with a vision language model for detecting and reporting anomalies. We experimentally demonstrate the effectiveness of PrimeGen across a variety of applications. PrimeGen can accommodate up to 955 amplicons, ensuring high amplification uniformity and minimizing dimer formation. Our development underscores the potential of collaborative agents, coordinated by generalist foundation models, as intelligent tools for advancing biomedical research.