Genomic Surveillance for Antimicrobial Resistance: A Bioinformatics Workflow

Genomic surveillance for antimicrobial resistance (AMR) uses whole-genome sequencing data to detect resistance determinants, track their spread, and support infection control decisions. This workflow guide covers the practical steps for building a reproducible AMR gene detection pipeline, from raw sequence inputs to interpreted reports. It is written for students, researchers, analysts, and life-science professionals who need concrete decisions about database selection, tool choices, quality thresholds, and result interpretation.

Whole-genome sequencing has become a standard approach for AMR surveillance in many settings, particularly in high-income countries, and innovations in sequencing and analysis technologies continue to reshape the field [5]. However, routine adoption of genomic methods faces real constraints, including bioinformatics capacity, standardization gaps, and interpretation challenges [13]. The workflow described here addresses those constraints directly by focusing on what you can control: input quality, database selection, validation steps, and documentation.

At a Glance

The table below summarizes the core decisions in an AMR genomic surveillance workflow. Each row links a workflow stage to the practical choice you must make and the record you should keep.

Scope and Purpose of Genomic AMR Surveillance

Genomic surveillance answers five core questions about resistance in a population: how large the problem is, whether resistance is increasing, whether new resistance types are emerging, whether a specific resistance type is spreading, and whether a particular resistance is linked to an outbreak [9]. These questions shape every downstream decision in the workflow, from sampling design to database choice.

The value of genomic methods lies in their resolution. Whole-genome sequencing can identify resistance determinants, virulence factors, and mobile genetic elements in a single assay [13]. This resolution supports outbreak investigations, transmission tracking, and the detection of emerging resistance mechanisms that phenotypic testing might miss. For example, genomic analysis of clinical Nocardia isolates revealed species-level resistance differences and strong correlations between specific genes and resistance phenotypes, information that would be difficult to obtain from susceptibility testing alone [6].

Genomic surveillance does not replace phenotypic susceptibility testing. Phenotypic methods remain necessary for detecting emerging resistant bacteria, new resistance mechanisms, and resistance trends [9]. The two approaches are complementary. Genomic data can predict resistance genotypes, while phenotypic data confirms whether those genotypes are expressed in the organism under standard testing conditions.

Data Inputs and Preparation

Sequence Data Types

The workflow accepts two primary input types: raw sequencing reads and assembled genomes. Raw reads, typically FASTQ files from Illumina or similar platforms, preserve the most information but require more compute for quality trimming and assembly. Assembled genomes, usually FASTA files, are smaller and faster to analyze but can hide assembly errors that affect resistance gene detection.

For surveillance programs that compare isolates across time or regions, consistency in input type matters more than the choice itself. A program that mixes raw reads for some samples and assemblies for others will produce results that are difficult to compare. Document the input type for every sample and keep that metadata with the analysis outputs.

Quality Thresholds

Read quality directly affects resistance gene detection accuracy. Low-quality reads produce truncated alignments and false-negative results. The following checks should be applied before any resistance gene analysis:

• Per-base quality scores across the read length
• Adapter contamination levels
• Estimated coverage depth relative to the expected genome size
• Assembly statistics including contig count and N50 value

Coverage depth is a critical threshold. Higher coverage supports confident detection of genes present in single copies and improves the resolution of mobile genetic element context. Low coverage can cause genuine resistance genes to be missed entirely. The exact threshold you choose depends on your sequencing platform and analysis goals, but the threshold must be documented and applied consistently across all samples in a surveillance run.

Metadata Capture

Genomic data is only as useful as its metadata. Each sample should carry identifiers for collection date, geographic origin, source type, host species, and clinical or environmental context. The NIH Genomic Data Sharing Policy outlines expectations for data sharing and metadata completeness in federally funded research [3]. Even outside funded programs, adopting similar metadata standards improves the long-term value of your surveillance data.

Metadata also supports the FAIR principles of findability, accessibility, interoperability, and reuse [4]. A resistance gene prediction without collection context cannot answer questions about spread or emergence. Build metadata capture into the workflow at the point of sample submission, not after analysis is complete.

Database Selection for Resistance Gene Detection

Major Database Categories

Antimicrobial resistance gene databases differ fundamentally in the number and type of genes they contain, the annotation structure, and the metadata stored for each entry [18]. No single database covers all resistance determinants, and the choice of database changes the results you obtain.

The most widely used databases include:

• Comprehensive Antibiotic Resistance Database (CARD), which organizes genes by resistance mechanism and includes detection models based on protein homology
• NCBI AMRFinderPlus, which integrates resistance gene detection with stress response and virulence gene identification
• ResFinder, which focuses on acquired resistance genes and includes species-specific variant detection
• MEGARes, which is designed for metagenomic data and includes structured ontologies for resistance mechanisms

Each database has strengths and limitations. CARD provides detailed mechanistic annotation but has known limitations in its decision model, which can produce false predictions in specific alignment scenarios [22]. NCBI AMRFinderPlus benefits from integration with the broader NCBI data resources [2]. ResFinder is straightforward for clinical isolates but may miss novel variants.

Database Selection Criteria

Choose a database based on your specific surveillance question, not on habit or convenience. The following criteria should guide the decision:

• Organism coverage: Does the database include the species in your surveillance program?
• Resistance mechanism coverage: Does it include the gene families relevant to your setting?
• Annotation depth: Does it provide mechanism class, substrate specificity, and mobility context?
• Update frequency: How often is the database updated with new resistance determinants?
• Detection model: Does it use BLAST-based homology, hidden Markov models, or a hybrid approach?

For surveillance programs tracking emerging resistance, database update frequency is a critical factor. Novel resistance genes continue to be discovered in environmental and clinical settings. One study identified a new integron-borne aminoglycoside resistance gene in clinical pathogens by screening environmental bacterial communities, a gene that had escaped discovery in sequenced clinical isolates for years [21]. A static database would miss such determinants entirely.

Multi-Database Confirmation

Run resistance gene detection against at least two independent databases and compare the results. Discrepancies between databases flag predictions that need manual review. This confirmation step is especially important for genes with high sequence similarity to non-resistance homologs or for predictions based on partial alignments.

The choice of confirmatory database should be documented in the analysis protocol. For example, a primary screen with CARD and a confirmatory screen with NCBI AMRFinderPlus provides complementary detection models and annotation structures. The comparison step should record which genes were detected by both databases, which were detected by only one, and the alignment metrics for each prediction.

Analysis Tools and Workflow Architecture

Tool Categories

The bioinformatics tools for AMR surveillance fall into five functional domains: pathogen identification, molecular epidemiology, resistance gene detection, virulence profiling, and mobile genetic element analysis [13]. A complete workflow integrates tools from each domain, though the specific combination depends on your surveillance objectives.

For resistance gene detection specifically, the main tool categories are:

• Command-line tools for local installation, which offer flexibility and reproducibility
• Web-based platforms, which reduce local compute requirements but may limit batch processing
• Cloud-based analysis platforms, which provide user-friendly interfaces and require minimal bioinformatics expertise [16]

Local Versus Cloud-Based Analysis

Local installation gives you full control over tool versions, parameters, and data security. It requires bioinformatics expertise, compute infrastructure, and ongoing maintenance. Cloud-based platforms reduce these barriers but introduce questions about data governance, cost, and output consistency.

A comparative review of three cloud-based platforms found that all were easy to use and offered pathogen identification, resistance gene detection, and epidemiological analysis [16]. However, the number of resistance genes detected varied widely across platforms for the same input data, from 83 to 358 genes across 90 isolates. This variation matters for surveillance programs that compare results across time or sites. If you use a cloud platform, document the platform version and analysis date for every run.

Workflow Reproducibility

Reproducibility requires version control for both tools and databases. Record the following for every analysis run:

• Tool name and version
• Database name and version or retrieval date
• Alignment parameters and thresholds
• Reference genome or assembly version
• Operating system and compute environment

Containerization tools such as Docker or Singularity can package the entire analysis environment, ensuring that the same workflow produces the same results regardless of where it runs. Workflow managers such as Snakemake or Nextflow add dependency tracking and automatic rerun capabilities.

Step-by-Step Analysis Workflow

Step 1: Quality Control and Trimming

Start with raw sequencing reads and apply quality trimming to remove low-quality bases and adapter sequences. Record the number of reads before and after trimming, the total bases retained, and the mean quality score. This step prevents spurious alignments that can produce false resistance gene predictions.

Step 2: Assembly or Read-Based Analysis

Decide whether to assemble reads into contigs or analyze reads directly. Assembly is appropriate for isolate sequencing where you expect a single organism. Read-based analysis is appropriate for metagenomic samples or when assembly quality is poor.

For assemblies, record the assembler version, assembly parameters, and quality metrics including N50 and total assembly length. For read-based analysis, record the read mapping approach and the reference database used.

Step 3: Species Confirmation

Confirm the species identity of each isolate before interpreting resistance gene predictions. Species misidentification is a known problem in clinical microbiology, and genomic methods can reclassify isolates that were misidentified by phenotypic or mass spectrometry methods [6]. Average Nucleotide Identity analysis is the standard approach for species confirmation.

Species confirmation matters because resistance genes have species-specific relevance. A gene that confers resistance in one species may be a poor predictor of resistance in another. The Klebsiella pneumoniae species complex illustrates this challenge, as its diversity and complex population structure complicate the interpretation of genome data [7].

Step 4: Resistance Gene Detection

Run the primary resistance gene detection tool against your chosen database. Record the following for each predicted gene:

• Gene name and allele or variant identifier
• Alignment identity and coverage
• Detection model score or expectation value
• Predicted resistance mechanism and substrate

Apply the quality thresholds defined in your protocol. Predictions that fall below the identity or coverage threshold should be flagged as low confidence and excluded from downstream analysis unless manually reviewed.

Step 5: Mobile Genetic Element Context

Resistance genes on mobile genetic elements pose a greater public health risk than chromosomal genes because they can spread horizontally across species. Plasmid typing and insertion sequence analysis provide the mobility context for resistance gene predictions [8].

The integration of resistance gene mobility into surveillance is an emerging priority. Current environmental surveillance often overlooks the significance of resistance gene mobility, which limits risk assessment accuracy [19]. For clinical and public health decision-making, the distinction between a chromosomal resistance gene and a plasmid-borne gene is essential.

Step 6: Phylogenetic Context

Place isolates in a phylogenetic context to identify clusters of closely related strains and track transmission. Core genome single nucleotide polymorphism analysis is the standard approach for outbreak investigation [10]. The choice of typing scheme affects the resolution and comparability of results.

For Neisseria gonorrhoeae surveillance, whole-genome sequencing with core genome multilocus sequence typing provided more consistent phylogenetic clusters than traditional typing schemes [11]. This finding supports the use of whole-genome approaches for surveillance programs that need high-resolution strain tracking.

Step 7: Interpretation and Reporting

Interpret resistance gene predictions in the context of the organism, the resistance mechanism, and the clinical or public health question. A gene prediction is not a resistance phenotype. The correlation between genotype and phenotype varies by species and gene family.

The report should include:

• Sample identifiers and metadata
• Species confirmation results
• Resistance gene predictions with quality metrics
• Mobile genetic element context
• Phylogenetic placement
• Interpretation notes and limitations
• Escalation criteria for clinical or public health action

Validation and Quality Control

Genotype-Phenotype Concordance

Validate resistance gene predictions against phenotypic susceptibility data whenever possible. The concordance between predicted and observed resistance provides a direct measure of workflow accuracy. Strong genotype-phenotype correlations have been documented for specific gene families, including sul1 in sulfamethoxazole and trimethoprim resistant Nocardia farcinica and bla AST-1 in beta-lactam resistant isolates [6].

Discrepancies between genotype and phenotype require investigation. Possible explanations include:

• Gene expression differences not captured by DNA sequence data
• Resistance mechanisms not represented in the database
• Phenotypic testing errors
• Sequence quality issues

Record the concordance rate for each surveillance run and track changes over time. A declining concordance rate may indicate emerging resistance mechanisms that your database does not cover.

False Positive and False Negative Sources

Repeat masking is a known source of bias in resistance gene prediction. Masking repetitive elements before gene prediction can remove genuine resistance genes that share sequence similarity with repeats [23]. If your workflow includes repeat masking, verify that known resistance genes in your control samples are still detected.

False positives arise from partial alignments, chimeric assemblies, and database entries with incomplete annotation. False negatives arise from low coverage, short reads that do not span the full gene, and novel variants with low identity to database sequences.

Control Samples

Include positive and negative control samples in every analysis batch. Positive controls are isolates with well-characterized resistance genes. Negative controls are isolates with no known resistance determinants. The controls validate that the workflow detects expected genes and does not produce spurious predictions.

Record control results in the analysis log. A workflow that fails to detect a known resistance gene in a positive control requires troubleshooting before sample results can be trusted.

Records and Measurements

Required Records

Maintain the following records for each surveillance run:

• Sample manifest with metadata
• Sequencing quality metrics
• Assembly statistics
• Database versions and retrieval dates
• Tool versions and parameters
• Resistance gene predictions with quality scores
• Genotype-phenotype concordance data
• Control sample results
• Interpretation notes and reviewer comments

Measurement Frequency

The frequency of surveillance measurements depends on the program objective. Outbreak investigation requires rapid turnaround and continuous monitoring. Trend surveillance requires consistent sampling intervals over time. The design of the surveillance system should specify the measurement frequency before data collection begins [9].

Data Sharing and Governance

Genomic surveillance data has value beyond the immediate analysis. Data sharing enables cross-regional comparisons and the detection of emerging threats. The NIH Genomic Data Sharing Policy provides a framework for responsible data sharing that balances openness with privacy protections [3].

Equitable data sharing and governance frameworks are essential for the successful implementation of genomic surveillance innovations [5]. Data sharing agreements should specify what data will be shared, under what conditions, and with what attribution requirements.

Common Failure Patterns

Database Version Drift

Using different database versions for different samples in the same surveillance program produces incomparable results. A gene detected with database version 1 may be absent from database version 2, or vice versa. Freeze the database version for the duration of a surveillance cycle and document the version in every report.

Threshold Inconsistency

Applying different identity or coverage thresholds to different samples biases the results. A sample analyzed with a 90 percent identity threshold will report fewer genes than one analyzed with an 80 percent threshold. Define thresholds in the protocol and apply them uniformly.

Metadata Loss

Samples that lose their metadata cannot be interpreted. A resistance gene prediction without collection date, source, or geographic origin has limited surveillance value. Build metadata validation into the sample intake process.

Overinterpretation of Gene Presence

Reporting a resistance gene as clinical resistance without phenotypic confirmation overstates the certainty of the finding. Gene presence indicates resistance potential, not resistance expression. The report should distinguish predicted resistance from confirmed resistance.

Compute Infrastructure Failure

Local analysis pipelines fail when compute infrastructure is inadequate. The simulation-based evaluation of a WHO-recommended workflow in Pakistan identified workforce training and bioinformatics infrastructure as limiting factors for implementation [17]. Plan for compute capacity before the workflow goes live.

Limitations and Interpretation Boundaries

Database Coverage Gaps

All resistance gene databases have coverage gaps. Novel resistance genes continue to be discovered, and databases lag behind the actual diversity of resistance determinants in clinical and environmental settings [21]. A negative result from a database search does not prove the absence of resistance.

Species-Specific Interpretation

Resistance genes have species-specific relevance. The same gene may confer high-level resistance in one species and have no phenotypic effect in another. Species confirmation is a prerequisite for meaningful interpretation.

Metagenomic Complexity

Metagenomic samples contain DNA from multiple organisms, complicating the assignment of resistance genes to specific hosts. The abundance of resistance genes in metagenomic data can be used as an indicator for surveillance, but the resolution is lower than for isolate sequencing [12].

Phenotypic Testing Remains Necessary

Genomic methods do not replace phenotypic susceptibility testing. Phenotypic methods remain necessary for detecting emerging resistant bacteria, new resistance mechanisms, and resistance trends [9]. The future of surveillance lies in integrating both approaches, not replacing one with the other.

Safety and Regulatory Context

Data Security

Genomic data from clinical isolates may contain information that requires protection under privacy regulations. The NIH Genomic Data Sharing Policy outlines expectations for data security and access controls [3]. Local regulations may impose additional requirements.

Laboratory Safety

Whole-genome sequencing workflows involve handling of cultured pathogens. Standard laboratory safety protocols for the organisms in your surveillance program apply. The workflow described here is a bioinformatics analysis and does not change the biosafety requirements for sample handling.

Responsible Use of Analysis Tools

The use of large language models and automated analysis tools in microbial genomics requires attention to reliability, reproducibility, and risk-aware interpretation [14]. Automated tools can accelerate analysis but cannot replace human review of critical findings.

Professional Escalation Criteria

Escalate results to clinical or public health authorities when the following conditions are met:

• Detection of a resistance gene with known clinical significance that was not previously present in the surveillance population
• Detection of a resistance gene on a mobile genetic element in a high-risk clinical isolate
• Genotype-phenotype discordance that suggests a novel resistance mechanism
• Phylogenetic clustering that suggests nosocomial or community transmission
• Detection of resistance to last-line antimicrobial agents

The escalation pathway should be defined in the surveillance protocol before the workflow goes live. The report should include the specific criteria that triggered escalation and the recommended next steps.

Frequently Asked Questions

What is the minimum sequencing coverage for reliable resistance gene detection?

Coverage requirements depend on the sequencing platform and the analysis goal. Higher coverage supports confident detection of single-copy genes and improves resolution of mobile genetic element context. The WHO-recommended workflow includes coverage thresholds as part of its quality metrics [17]. Document your coverage threshold and apply it consistently across all samples.

Which resistance gene database should I use for clinical surveillance?

The choice of database depends on your organism, resistance mechanisms of interest, and analysis infrastructure. CARD provides detailed mechanistic annotation, NCBI AMRFinderPlus integrates with broader NCBI resources [2], and ResFinder focuses on acquired resistance genes. Run at least two databases and compare results to flag predictions that need manual review.

How do I validate resistance gene predictions?

Validate predictions against phenotypic susceptibility data whenever possible. The concordance between predicted and observed resistance provides a direct measure of workflow accuracy [6]. Include positive and negative control samples in every analysis batch and record control results in the analysis log.

Can genomic surveillance replace phenotypic susceptibility testing?

No. Genomic methods detect resistance genes, while phenotypic methods confirm resistance expression. Phenotypic testing remains necessary for detecting emerging resistant bacteria, new resistance mechanisms, and resistance trends [9]. The two approaches are complementary.

How do I handle discrepancies between genotype and phenotype?

Investigate discrepancies systematically. Possible explanations include gene expression differences, resistance mechanisms not represented in the database, phenotypic testing errors, and sequence quality issues. Record the concordance rate for each surveillance run and track changes over time.

What metadata should I collect for each sample?

Collect identifiers for collection date, geographic origin, source type, host species, and clinical or environmental context. The NIH Genomic Data Sharing Policy outlines expectations for metadata completeness [3]. Metadata supports the FAIR principles of findability, accessibility, interoperability, and reuse [4].

How do I detect novel resistance genes that are not in my database?

Novel resistance genes require approaches beyond database searching. Functional metagenomics followed by filtering of known resistance genes has successfully identified new resistance determinants in environmental samples [21]. These approaches require specialized expertise and are not part of routine surveillance workflows.

What are the main barriers to implementing genomic surveillance in low-resource settings?

The main barriers include bioinformatics capacity, standardization, and interpretability [13]. Workforce training and compute infrastructure are critical limiting factors [17]. Cloud-based platforms can reduce some barriers but introduce questions about cost, data security, and output consistency [16].

Related Bioinformatics Guides

• Predicting AMR from Genomic Data
• Machine Learning Classification of Antimicrobial Resistance Genes
• Computational Approaches to Understanding Antimicrobial Resistance (AMR)
• Genomic Selection in Animal Breeding
• Gene Ontology (GO) and Enrichment Analysis

References and Further Reading

• EMBL-EBI Training. European Bioinformatics Institute.
• NCBI Data Resources. National Center for Biotechnology Information.
• Genomic Data Sharing Policy. National Institutes of Health.
• The FAIR Guiding Principles. Scientific Data.
• Innovations in genomic antimicrobial resistance surveillance.. The Lancet. Microbe, 2023.
• Genomic and phenotypic characterization of antimicrobial resistance in clinical Nocardia species isolates.. Frontiers in cellular and infection microbiology, 2025.
• A genomic surveillance framework and genotyping tool for Klebsiella pneumoniae and its related species complex.. Nature communications, 2021.
• Genomic surveillance of antimicrobial resistance in bovine fecal samples from Lebanon.. Veterinary microbiology, 2025.
• Present and Future Surveillance of Antimicrobial Resistance in Animals: Principles and Practices.. Microbiology spectrum, 2018.
• Genomic epidemiology and antimicrobial resistance of Corynebacterium macclintockiae, the predominant species of human pathogens within the Corynebacterium jeikeium complex.. Journal of clinical microbiology, 2025.
• Genomic trends and emerging antimicrobial resistance in Neisseria gonorrhoeae over two decades in Kenya.. Microbiology spectrum, 2025.
• Genomic evolution of antimicrobial resistance in Escherichia coli.. Scientific reports, 2021.
• Strategies for integrating whole-genome sequencing into antimicrobial resistance surveillance.. 2026.
• Responsible Use of Large Language Models in Microbial Genomics and Bioinformatics: A Life-Science Framework for Reliability, Reproducibility, and Risk-Aware Interpretation. 2026.
• Application value and challenges associated with plasma cell-free DNA metagenomic sequencing technology in the diagnosis of infections in patients with hematological disorders. 2026.
• A comparative review of three cloud-based platforms for microbial whole genome sequencing analysis.. 2026.
• Whole-Genome Sequencing for Antimicrobial Resistance Detection: A Simulation-Based Evaluation of WHO-Recommended Workflow in Pakistan.. 2026.
• Review and Comparison of Antimicrobial Resistance Gene Databases. Antibiotics, 2022.
• Towards the integration of antibiotic resistance gene mobility into environmental surveillance and risk assessment. npj Antimicrobials and Resistance, 2025.
• A Comprehensive Self-Resistance Gene Database for Natural-Product Discovery with an Application to Marine Bacterial Genome Mining. International Journal of Molecular Sciences, 2023.
• Discovery of a novel integron-borne aminoglycoside resistance gene present in clinical pathogens by screening environmental bacterial communities. Microbiome, 2020.
• Deep analysis and optimization of CARD antibiotic resistance gene discovery models. BMC Genomics, 2019.
• Bias in resistance gene prediction due to repeat masking. Nature Plants, 2018.
• Genome-wide characterization of sugarcane catalase gene family identifies a ScCAT1 gene associated disease resistance.. International Journal of Biological Macromolecules, 2023.
• Enhancing clinical microbiology for genomic surveillance of antimicrobial resistance implementation in Africa. Antimicrobial Resistance and Infection Control, 2024.
• Genomic determinants of antimicrobial resistance in surveillance systems: from detection to decision-relevant interpretation for clinical and public health action. Discover Applied Sciences, 2026.
• Antimicrobial Resistance as a Worldwide Crisis and the Role of Genomic Surveillance in Monitoring and Combating It: A Comprehensive Review. Bacteria, 2026.
• Genomic insights into antimicrobial resistance and cytotoxin production in Klebsiella oxytoca complex isolates from a nationwide Japanese surveillance program. Journal of Global Antimicrobial Resistance, 2026.
• Confronting the Antimicrobial Resistance Crisis in China: Emerging Superbugs, Genomic Surveillance, and Innovative Countermeasures. China Cdc Weekly, 2025.

This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.