This skill drives postdoc-grade antimicrobial-resistance surveillance from bacterial WGS. It covers acquired AMR gene detection (AMRFinderPlus, ResFinder 4.0, CARD-RGI, abricate), chromosomal point-mutation calling (PointFinder, AMRFinderPlus species mode, TB-Profiler, Mykrobe), mobile-genetic-element context (MOB-suite, PlasmidFinder, MobileElementFinder), and phenotype prediction against EUCAST / CLSI breakpoints. Cross-tool reporting is harmonised through hAMRonization to the PHA4GE schema, and TB drug-resistance prediction is interpreted against the WHO 2nd-edition catalogue. The skill is decision-grade: it tells the agent when to switch from AMRFinderPlus to TB-Profiler, how to handle WHO Group 3 (Uncertain) calls without misleading clinicians, when heteroresistance demands deep read-based variant calling, and when to insist on long-read assembly for plasmid resolution.
Distinct from metagenomics/amr-detection (community-level ARG quantification, not isolate-focused) and from variant-calling/variant-calling (which owns SNP-calling mechanics that this skill consumes).
conda install -c bioconda ncbi-amrfinderplus resfinder rgi abricate hamronization abritamr staramr tb-profiler mykrobe mob_suite plasmidfinder mefinder lofreq snippy
amrfinder -u
tb-profiler update_tbdb
pip install pandas biopython
For long-read or hybrid assembly upstream:
conda install -c bioconda flye unicycler medaka
Tell the AI agent what surveillance question to answer:
"We have 250 Illumina paired-end runs of clinical E. coli and Klebsiella pneumoniae from a tertiary hospital ICU surveillance project spanning Q1-Q4 2025. Assemble with Unicycler, run AMRFinderPlus with
--organismset per isolate plus--plus, run ResFinder 4.0 with--acquired --point -s 'species', harmonise via hAMRonization, and produce a long-format CSV mapping isolate -> antibiotic class -> highest-confidence determinant -> predicted phenotype, in WHO GLASS reporting categories (S / I / R / Insufficient evidence) against EUCAST 2025 breakpoints. Document every tool version and database date."
"Twenty Klebsiella pneumoniae isolates from a suspected NDM-1 outbreak across two wards. For each, identify the carbapenemase via AMRFinderPlus, reconstruct plasmid contigs via MOB-suite
mob_recon, type each plasmid viamob_typer, check for IS26 / ISKpn7 mobilisation via MobileElementFinder, and tell me whether the NDM-1-bearing plasmid is the same replicon and MOB-suite cluster across isolates. Recommend a long-read confirmation isolate if plasmid context appears fragmented in any short-read assembly (Method=PARTIAL_CONTIG_ENDin AMRFinderPlus output)."
"Sixteen MTB clinical assemblies from a high-burden setting. Run TB-Profiler with the current WHO 2nd-edition catalogue; report per-drug resistance call (R / R-interim / Uncertain / S / Not predicted) for the 13 anti-TB drugs in the catalogue, flag Coll/Napier lineage assignment, and call out high-confidence rpoB / katG / pncA / gyrA / inhA mutations explicitly. Run Mykrobe as a cross-check on any R or XDR call. Do NOT collapse Group 3 (Uncertain) mutations to 'S' -- report them as 'phenotypic DST recommended'."
"Five Klebsiella pneumoniae clinical isolates with treatment failure on colistin. Run deep read-based variant calling with lofreq at MAF >= 1% across mgrB / pmrAB / phoPQ, report any non-reference allele frequency between 1% and 20% as heteroresistance candidates, and compare against the assembly-based AMRFinderPlus output. For any heteroresistance candidate flag for targeted deep amplicon sequencing."
"Hundred isolates from a one-year colistin-resistance surveillance project across food-animal and clinical sources. Detect mcr-1 through mcr-10 acquired genes via AMRFinderPlus; for each mcr-9 and mcr-10 hit, flag that phenotypic colistin MIC confirmation is required before reporting as colistin-R. Cross-reference with MOB-suite to determine plasmid host-range. Produce a per-sample table: mcr variant, plasmid replicon, predicted host range, phenotypic confirmation status."
"I have AMRFinderPlus calling
bla_NDM-5, ResFinder callingbla_NDM-1, and RGI callingbla_NDM-1. Pass everything through hAMRonization, reconcile against NCBI ReferenceGeneCatalog nomenclature, and tell me whether the discordance is real allele difference or database-naming drift. If real, explain the MIC implications."
--organism for E. coli / Salmonella / Klebsiella / Staphylococcus / etc.; TB-Profiler for M. tuberculosis; ResFinder 4.0 or abritAMR for phenotype prediction.mob_recon + mob_typer and recommend long-read confirmation if short-read assembly produces fragmented plasmid contigs (Method=PARTIAL_CONTIG_END).hamronize with mandatory PHA4GE provenance metadata, then hamronize summarize for the cohort table.amrfinder -n mtb.fa returns an empty AMR table on an XDR-TB genome. Use TB-Profiler.tb-profiler list_db for the bundled edition before clinical handover.Method=PARTIAL_CONTIG_END in AMRFinderPlus output is the smoking gun for plasmid-context fragmentation in short-read assembly. Re-assemble with long-read (Nanopore R10.4.1 Q20+ or PacBio HiFi) or hybrid before reporting MGE context.--organism Klebsiella_pneumoniae suppresses intrinsic-gene noise (fosA, oqxAB). Cross-organism comparisons that mix --organism-on with --organism-off output are not pooled-comparable.-q 13 -a 0.01.