AI News

DNA sequencing pipelines already automate execution through workflow systems, but operators still make manual decisions about sample-specific quality thresholds, borderline variant calls, anomalies, and expert review. The paper introduces BaseCamp, an agentic AI framework focused on this decision layer. It divides the work among six specialized agents for sample intake and quality control, alignment, variant calling, annotation, cross-stage monitoring, and reporting. The agents do not perform sequence analysis themselves: established bioinformatics tools handle execution, while the agents select and configure tools, interpret results, and determine the next step. A consortium of fine-tuned, domain-specialized large language models coordinated by a central reasoning LLM powers the system locally, so sequencing data remains in the operating environment, with human-in-the-loop orchestration. In evaluation, agent-generated configurations were concordant with expert practice, an explicit filtering ledger made removed information inspectable, and cross-stage anomaly detection found conditions missed by execution monitoring. The authors present BaseCamp as a potentially generalizable design for agentic automation of scientific data pipelines.