Nanopore Metagenomics Benchmark

AMR Species Attribution

Resistance genes were detected in the evaluation pool, but could the reads be named to the right species? Kraken2's call for each supporting read was compared with the true organism from the supplier genomes. The registered 0.85 target was not met, and the report shows where it fell short.

0.593Reads named to the correct species (51 of 86) — registered target 0.85
22/22cepA reads attributed correctly
16/16tet(W) reads attributed correctly
13/48tet(Q) reads attributed correctly (0.271)

Where each supporting read was placed

Each bar is one true organism and gene, using the supplier genomes as truth. Hover a segment for its read count.

Show: Gene:
Correct species Right lineage, not to species Wrong organism

Total 86 reads with a true-organism label. Of 91 supporting reads, 89 fall in evaluable families and 3 had a weak, ambiguous or missing alignment to the supplier genomes, so they were not scored.

Overall share of reads named to the correct species

Registered prediction E2 against the observed value, with the other three outcome shares.

Correct species
0.593
Right lineage, not to species
0.314
Wrong organism
0.093

Registered predictions (Addendum A8, evaluation pool)

#PredictionObservedOutcome
E1All four species × gene units attributed4 of 4Met
E2Correct species ≥ 0.85 overall0.593Not met
E3Wrong organism ≤ 0.100.093Met
E4Cross-carrier swaps among tet(Q) reads ≤ 10Met
E5No false organism call (≥ 3 reads, non-carrier species)0Met

Why the target was missed. Most Kraken2 k-mer windows of a nanopore read have no hit. The tet(Q) and tet(W) gene sequences map only to a shared ancestor, while cepA maps to the species, so a read reaches its species only if it carries a species-specific window elsewhere (Addenda A9 and A10). The tet(Q) loci of B. fragilis and P. corporis also share about 20 kb at about 95% identity.

Mock community only (17 species, 21 strains): the true organism comes from the supplier genomes, which a real sample does not have, so this is a worked example and not an accuracy estimate. Research use only, not clinically validated.

View the repository on GitHub →