Bio Metagenomics Amr Detection

Version Compatibility

Reference examples tested with: AMRFinderPlus 3.12+, RGI 6+ (CARD 3.2+), ABRicate 1.0+, pandas 2.2+.

Before using code patterns, verify installed versions match. If versions differ:

• CLI: amrfinder -V (reports software AND database version), rgi main --version, abricate --list to confirm DB snapshots
• Python: pip show <package> then help(module.function) to check signatures

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

The AMR reference DATABASE is versioned and updated roughly monthly; amrfinder -V reports both software and database version, and ABRicate ships pinned database snapshots so two labs on different versions get different calls. Record the tool version, the database version, and (for read-based work) the normalization unit and sequencing depth - none are recoverable later.

AMR Detection (Community Resistome)

"What resistance genes are in my community, and how abundant?" -> Match reads or contigs to a curated ARG database - reporting that ARG sequences are present at some relative abundance, never that the sample is resistant, because a metagenomic hit has no host and no expression.

• CLI (reads): rgi bwt -1 R1.fq.gz -2 R2.fq.gz -a kma -n 16 -o sample --local
• CLI (contigs/MAGs): amrfinder -n contigs.fasta --plus -o amr.tsv

Scope: community/metagenomic resistome - read-based quantification and contig/MAG presence calling. Pure-culture isolate AMR, point-mutation resistance, in-silico antibiogram, MLST/clone/outbreak context, and GLASS reporting -> epidemiological-genomics/amr-surveillance. Assembly/binning and ARG-host linkage mechanics -> genome-assembly/metagenome-assembly. General gene-family/pathway abundance -> functional-profiling.

The Single Most Important Modern Insight -- An ARG Hit Is a Sequence Match, Not a Phenotype

An ARG hit is a match against a reference database, not a measured resistance phenotype - and in a metagenome it is a match with no host and no genomic context until assembly. Three inferences a naive pipeline silently makes, all wrong:

1. match -> gene - defeated by partial hits (a 30 bp conserved-domain hit to a 1 kb ARG is not a gene); guarded by gene-fraction / breadth-of-coverage.
2. gene -> resistance - defeated by expression and regulation: silent sul2 below ECOFF, ampC/blaOXA driven only when an IS lands in the promoter, efflux that needs overexpression, truncations that still score a partial hit, point mutations where only the SNP matters.
3. gene -> who carries it / is it mobile - defeated by read shortness: a short read cannot see its neighbors, so host and plasmid context need assembly, long reads, or Hi-C.

The honest deliverable is "these ARG sequences are present at this relative abundance in this community," never "this sample is resistant to drug X." The moment a report says resistant, it has smuggled in a host, an expression assumption, and a clinical breakpoint the data never contained. On a pure culture the organism can be grown and an MIC measured - that is a different skill (epidemiological-genomics/amr-surveillance).

Read-Based vs Assembly-Based: the Core Metagenomic Tradeoff

The assembly paradox: metagenomic assemblies preferentially break exactly at ARG/MGE boundaries, recovering only a small fraction of true ARG genomic contexts and underestimating the resistome (Abramova 2024 BMC Genomics 25:959). So "assemble to get host" is necessary but not sufficient - long reads (Nanopore/PacBio) and Hi-C metagenomics are the real remedy for ARG-host/MGE linkage.

Tool Taxonomy

Decision Tree by Scenario

Read-Based Resistome Quantification

# CARD read mapping (homolog models). RGI bwt CANNOT screen point-mutation SNPs, so this is for
# acquired/homolog ARGs only - never report a gyrA read hit as fluoroquinolone resistance.
rgi bwt -1 reads_R1.fq.gz -2 reads_R2.fq.gz \
    -a kma -n 16 \
    -o sample_resistome --local
# Outputs *.gene_mapping_data.txt with percent coverage and depth per gene.

# AMR++/MEGARes applies the gene-fraction filter (default 80%): the minimum proportion of a
# reference covered by >=1 read for "present" - the read-based analog of breadth-of-coverage.

ARGs-OAP/SARG normalizes to copies-of-ARG-per-16S or per-cell; report the unit. Gene fraction (breadth) is the single most important false-positive guard - without it a conserved-domain fragment counts as a present gene.

Contig / MAG Presence Calling

amrfinder -n contigs.fasta \
    --plus \                  # also report biocide/metal (STRESS) and virulence elements
    --threads 8 -o amr.tsv
# --ident_min default -1 = use the per-gene CURATED cutoffs; overriding with a global value is usually a mistake.
# Point mutations require --organism (a single known species) - inappropriate for a mixed community;
# use it only on a taxonomically resolved MAG, and defer isolate point-mutation work to amr-surveillance.

AMRFinderPlus uses manually curated per-gene BLAST cutoffs (plus HMM cutoffs with protein), not a flat 80/80 - catching divergent real variants while rejecting partial housekeeping homologs. ABRicate, by contrast, is flat 80/80 and acquired-genes-only; it will never report a point mutation.

Per-Method Failure Modes

ARG presence reported as resistance

Trigger: an output column or summary that says "resistant." Mechanism: presence is not expression and not a host-linked MIC. Symptom: a sewage metagenome described as "resistant to carbapenems." Fix: report "ARG detected at abundance X"; reserve phenotype claims for isolates (amr-surveillance).