Instructing a Chatbot to Design Nucleic Acid Probes for Diagnostics

Large language models (LLMs) with their natural language processing and automation capabilities can streamline the design of nucleic acid diagnostic assays, but they can produce inaccurate outputs. Here, we developed an LLM-based automated design system for designing quantum dot barcode (QDB) and polymerase chain reaction (PCR) assays with high accuracy. We leveraged a structured prompting approach that combines domain knowledge, task planning, and tool use instructions. We used this system to design QDB assays for 1,512 genomes of infectious viruses in 24 h. We showed the versatility of this LLM system for generating PCR primers for infectious pathogens. We validated the QDB assay designs against 7 viruses with high epidemic-causing potential. Those QDB assays paired with the corresponding PCR products exhibit high analytical sensitivity (10 copies/μL). They also exhibited high specificity in multiplex respiratory and viral hemorrhagic fever panels. Our approach allows us to rapidly design nucleic acid diagnostics in epidemics or pandemics.