A Pipeline for Mapping FDA-Approved Nucleic Acid Amplification Tests (NAATs) to NCBI Taxonomy

Background The United States Food and Drug Administration (FDA) catalogs in vitro diagnostics, including Nucleic Acid Amplification Tests (NAATs), using unstructured clinical terminology. Because these regulatory labels often diverge from standardized biological taxonomy, they obstruct automated post-market surveillance and computational integration, particularly for in silico inclusivity audits against emerging pathogen variants. This study aimed to build a reproducible computational pipeline that translates historical FDA microbial NAAT records into machine-actionable biological identifiers. Methods Synthesizing clinical diagnostic logic, taxonomic normalization rules, and modern text mining, the workflow utilized the R ellmer package to interface with a Large Language Model (LLM) for Technology Classification, successfully isolating true NAATs from 510(k) summary documents. Subsequent algorithmic partitioning and LLM-driven Named Entity Recognition (NER) mapped complex syndromic labels and clinical abbreviations to National Center for Biotechnology Information (NCBI) Taxonomy Identifiers (TaxIDs). Results While only 20% of initial FDA records mapped directly to NCBI Taxonomy, the automated deconvolution of syndromic panels and NER extraction expanded and normalized the dataset. The pipeline successfully generated FDA510kPathogens, a highly granular master repository containing 457 unique pathogen-to-submission entries formally linked to enriched NCBI TaxIDs and metadata. Conclusions By programmatically translating human-readable regulatory metadata into machine-actionable biological descriptors, this open-source digital infrastructure resolves a critical informatics bottleneck. This provides diagnostic laboratories and developers with the foundational data necessary to conduct continuous, automated validation of molecular assays.