Sequora: Agentic Sequence Analysis for Biosecurity and Genome Curation

A fast-arising risk of artificial intelligence (AI) is the large-scale, democratized ability to design new-to-nature DNA, RNA, or protein sequences, with the added danger that such sequences could evade existing biosecurity screening. Conversely, AI can help address this risk with auditable, evidence-based analysis of these sequences and their putative function. We introduce Sequora, a specialized agentic system for analyzing sequence function and potential pathogenicity. Sequora uses a large language model (LLM) to choose among 70 bioinformatics tools exposed through six Model Context Protocol (MCP) servers, and it returns biological claims linked to supporting records with model-assigned confidence scores. On a benchmark of AI-redesigned proteins of concern and negative controls, Sequora achieved 90.7% sensitivity at 98.9% specificity (n = 2,151). Sensitivity exceeded all seven configurations of the four screening providers evaluated including versions patched after disclosure, with comparable specificity. Sequora also identified the reference parent protein for 84.9% of its true positives. For the WHO critical-priority pathogen Candidozyma auris, we assessed Sequora on 111 proteins that were previously unannotated in the most recent NCBI release. A professional Candida Genome Database curator rated 94.6% of Sequora annotations acceptable under a six-point rubric, versus 17.1% for the same LLM without tools (Wilcoxon p = 7.1e-18). In one case, Sequora linked a previously unannotated protein to the GWT1 family, an established antifungal target. The applications illustrate how specialized agentic sequence analysis can support biological screening and broader biological research.