qualibact v1.1 methods

qualibact-v1.1 is the active refinement track on top of qualibact-v1.0. The underlying engine, dataset, outlier detection, and statistical pipeline are identical to v1.0; v1.1 only captures per-species refinements layered on top — either additional reference genomes brought in beyond AllTheBacteria, or specific threshold revisions requested by subject-area experts after the Nov 2025 QualiBact assessment survey.

For the engine pipeline, dataset construction, outlier filtering, statistical comparisons, threshold-selection logic, and the PASS / WARN / FAIL verdict model the website applies to every scheme, see the qualibact-v1.0 methods page.

A species page falls back to its v1.0 thresholds whenever no v1.1 row is published for that species — so v1.1 only needs to ship rows for species where the bounds actually change.

Per-species refinements

Achromobacter xylosoxidans

The qualibact-v1.1 thresholds for A. xylosoxidans incorporate 441 additional genomes beyond the AllTheBacteria-2024-08 set used for v1.0: 171 from BioProject PRJNA1234112 and 76 from PRJNA1148967 (both 2025, not yet in AllTheBacteria), 108 not yet publicly released, and 69 assemblies pulled directly from NCBI. Curated with the Rigshospitalet (, UZ Brussel (, and Vilnius University ( groups.

Campylobacter coli

Re-run through qualibact-engine on a curated reference set (98 ATB assemblies + 92 RefSeq references). The v1.1 thresholds for C. coli are noticeably tighter than v1.0: N50 lower raised from 25 kb to 31 kb, contig-count upper tightened from 180 to 120, and GC-content / TCS / assembly-size bands all narrowed. The species page calls out where the bounds differ from v1.0.

Klebsiella variicola

Re-run through qualibact-engine on a v1.1-specific reference set with ML-adjusted (FINAL_LOWER / FINAL_UPPER) bounds. Same methodology as v1.0; the species page calls out where the bounds differ from v1.0.

Klebsiella quasipneumoniae

K. quasipneumoniae v1.1 currently inherits its reference distribution from an earlier engine run (the legacy ESGEM-AMR-v1 dataset) and is queued for a fresh engine re-run. The species page renders an amber banner noting this. Once the re-run lands, the thresholds will be derived consistently with the rest of v1.1.

Expert-feedback survey adjustments

The Nov 2025 QualiBact assessment survey gathered structured feedback from 39 respondents across 37 species. Overall confidence was high (median 9 / 10), but a number of species drew specific revision requests. Every adjustment below is implemented as a source: pinned override on top of the v1.0-rev2 engine output; the full audit trail (who asked for what, when, why) is the Hand-pinned threshold rationale table rendered at the bottom of this page.

Assembly size

• Campylobacter jejuni — upper raised from 2.0 Mb to 2.1 Mb (INSA Lisboa Campylobacter subgroup). 2.0 Mb was excluding a small number of well-supported, high-quality genomes drawn from a diverse human / food / animal / environmental isolate set.
• Neisseria meningitidis — upper tightened from 2.4 Mb to 2.3 Mb (INSA Lisboa N. meningitidis subgroup). The reference dataset is consistently below 2.3 Mb; the tighter bound flags potentially contaminated or chimeric assemblies that the looser v1.0 bound passed.
• Neisseria lactamica — upper loosened from 2.3 Mb to 2.4 Mb to accommodate species diversity (N. gonorrhoeae subgroup).
• Haemophilus influenzae — upper tightened from 2.2 Mb to 2.0 Mb (Haemophilus subgroup). Two RefSeq references (GCF_002985465.2, GCF_002984345.2) pushed the engine's automatic bound up despite inconclusive ANI (~94%) against H. influenzae; pinned at 2.0 Mb to flag those and any future contamination.
• Haemophilus haemolyticus — upper proposed at 1.75 Mb (Haemophilus subgroup). Note: 1.75 Mb currently sits below the engine's lower bound (1.8 Mb), so the PASS band collapses; flagged for follow-up with the expert before publication.

GC content

• Campylobacter fetus — lower loosened to 32 % (Campylobacter fetus subgroup). C. fetus subsp. venerealis carries a large accessory genome that can pull GC below the engine's v1.0-rev2 lower of 32.99 %; the pin retains venerealis isolates.

Contamination

• Neisseria bergeri, N. flavescens, N. lactamica, N. perflava, N. polysaccharea — upper pinned at 3.0 as a general threshold across commensal Neisseria (N. gonorrhoeae subgroupset analysis). Engine v1.0-rev2 emitted values between 1 and 4 across the group; 3.0 keeps the band consistent.

N50 / contig count

• Neisseria gonorrhoeae — N50 lower raised from ≈23 kb to 25 kb (N. gonorrhoeae subgroup, based on the distribution of their data).
• Haemophilus influenzae, H. haemolyticus — contig-count upper pinned at 100 (Haemophilus subgroup). Assemblies with >100 contigs are most often contaminated; engine v1.0-rev2 emitted 140 and 110 respectively.

Coding-sequence count

• Neisseria gonorrhoeae — TCS lower raised from 1800 to 1900 (N. gonorrhoeae subgroup, based on the WHO reference genomes).
• Neisseria lactamica — TCS band widened to 1800 – 2800 (engine emitted 1900 – 2300) to accommodate species diversity (N. gonorrhoeae subgroup).

Completeness

• Neisseria flavescens, Neisseria polysaccharea — Completeness lower pinned at ≥ 99 (N. gonorrhoeae subgroup). Engine v1.0-rev2 emitted 94 and 98; pinned at 99 for consistency across the commensal group.

Further per-species adjustments are tracked as additional respondents return specific values.

Input data for v1.1

Each per-species threshold draws on:

1. The aggregated AllTheBacteria assembly stats and RefSeq reference reports that qualibact-v1.0-rev2 ingests (see the v1.0 methods page for the dataset construction).
2. A per-species *_refseq_genomes.csv.xz (BioSample, assembly accession, ANI, annotation info) — the public copy lives under /static/species/{Species}/qualibact-v1.1/.
3. A per-species *_assembly_stats.csv.gz containing the per-assembly inputs (sample, sylph species call, N50, contig count, longest, total length, completeness, contamination, TCS, genome size, GC) that the engine used to derive the v1.1 distribution. The public copy lives alongside the refseq CSV.

Acknowledgements

The QualiBact assessment survey was answered by experts across 15 genera. The full list of contributors for each per-species adjustment is shown on the relevant species pages and aggregated on the contributors page.

Source code

github.com/cgps-group/qualibact (website) and github.com/cgps-group/qualibact-engine (engine).