CARD Database

The ARO is the central organizing principle of CARD, a controlled vocabulary that defines terms for AMR genes, mutations, mechanisms, and antibiotics. It integrates with the broader Open Biomedical Ontologies (OBO) foundry, enabling computational reasoning and standardized annotation. Key features include:

• Hierarchical structure: Links specific beta-lactamase variants to the broader 'beta-lactam resistance' class.
• Cross-referencing: Maps to external databases like GenBank, UniProt, and PubMed.
• Mechanistic classification: Categorizes resistance by molecular mechanism (e.g., antibiotic efflux, target alteration, enzymatic inactivation).

CARD moves beyond simple presence/absence calls by incorporating sophisticated statistical models:

• Protein Homolog Models: Position-specific scoring matrices (PSSMs) built from curated alignments of resistance protein families, enabling the detection of remote homologs.
• Protein Variant Models: Capture single nucleotide polymorphisms (SNPs) and small indels that confer resistance in essential genes (e.g., gyrase mutations causing fluoroquinolone resistance).
• rRNA Mutation Models: Specifically designed to identify resistance-conferring mutations in ribosomal RNA genes, which are critical for detecting aminoglycoside and macrolide resistance.

CARD:Live is a dynamic, continuously updated component that ingests new AMR sequences from public repositories like NCBI daily. It automatically:

• Clusters novel sequences and flags them for curator review.
• Assigns preliminary ARO terms using machine learning classifiers.
• Provides a real-time dashboard of emerging resistance threats, bridging the gap between static database releases and the rapid pace of microbial evolution. This system is critical for proactive public health surveillance.

The foundation of CARD's accuracy is its manually curated collection of reference sequences. Each entry undergoes rigorous biocuration:

• Literature Validation: Every gene and mutation is linked to peer-reviewed experimental evidence demonstrating its role in resistance.
• Expert Review: A global network of AMR experts continuously refines annotations and resolves conflicts.
• Metadata Richness: Entries include details on the antibiotic class affected, the resistance mechanism, and the clinical significance, providing context beyond the raw sequence.

WildCARD is a community-driven annotation platform that allows researchers to submit novel resistance determinants for inclusion in CARD. This crowdsourcing model accelerates discovery by:

• Enabling direct submission of sequences with supporting evidence.
• Facilitating peer review through a transparent, open process.
• Bridging the gap between individual research findings and a globally accessible, standardized resource. It ensures CARD remains comprehensive and responsive to the research community's needs.

A structured, hierarchical controlled vocabulary that forms the semantic backbone of CARD. ARO organizes resistance determinants into a directed acyclic graph with relationships including:

• is_a: Parent-child subclass relationships (e.g., OXA-1 is_a Class D beta-lactamase)
• confers_resistance_to: Links genes to specific antibiotic classes
• part_of: Connects components to larger complexes (e.g., efflux pump subunits)

Each term receives a unique ARO accession number, enabling machine-readable, ontology-driven analysis that supports automated reasoning and consistent annotation across bioinformatics pipelines.

CARD extends beyond binary presence/absence calls by incorporating resistome variant data that captures the natural diversity of resistance genes across sequenced isolates. Key features include:

• Prevalence data: Frequency of specific AMR alleles across global isolate collections, helping distinguish widespread resistance mechanisms from rare variants
• Allelic variants: Catalogued sequence diversity within resistance gene families, supporting precise epidemiological tracking
• SNP-level resolution: Curated point mutations in chromosomal targets (gyrA, rpoB, 23S rRNA) with documented MIC shifts

This variant-level curation enables high-resolution surveillance and outbreak tracing.

CARD employs a dual-threshold paradigm to balance sensitivity and specificity in AMR detection:

• Perfect Cut-off: Used for well-characterized resistance genes where any match above this threshold reliably indicates a functional resistance determinant. Minimizes false negatives for clinically critical genes like NDM, KPC, and MCR families
• Strict Cut-off: Applied to genes with close homologs in non-resistant contexts (e.g., housekeeping genes). Requires higher sequence identity to prevent false-positive resistance calls

These empirically derived thresholds are continuously refined through literature curation and functional validation studies.