BERTax Distantly related dataset: taxonomic classification at superkingdom and phylum (Mock et al. 2022)
Each tool classifies the distantly related test fragments using only that dataset's training partition; scored by macro AveP (Table 1), accuracy and proportion of samples classified (SI Table S2), and accuracy after removing test sequences with a BLAST hit to the training partition (SI Table S5).
Overview
An explanatory profile has not yet been reviewed for this record.
Consult the linked sources for architecture or protocol details. Missing evidence is not evidence of a missing capability.
9 recorded evaluations, 72 metric rows. A comparison chart has not yet been validated for these results. The table retains the individual findings and their sources.
Results
Results are available, but no reviewed comparison panel is linked in this release.
All evaluations
9 evaluations · 72 results. Different protocols are not a single leaderboard.
Protocol: BERTax Distantly related dataset: taxonomic classification at superkingdom and phylum (Mock et al. 2022) · Dataset: BERTax Distantly related dataset (Mock et al. 2022)
Sorted by Accuracy (phylum rank; among the samples the tool classified; SI Table S2 'acc.') (higher is better). The best value in each column is highlighted. Decimals are rounded for display; each value links to the printed value and its source.
| Tested configuration | Accuracy (phylum rank; among the samples the tool classified; SI Table S2 'acc.') | Accuracy (superkingdom rank; among the samples the tool classified; SI Table S2 'acc.') | Macro average precision (phylum rank; macro average over classes; unclassified samples treated as random guesses) | Macro average precision (superkingdom rank; macro average over classes; unclassified samples treated as random guesses) | Accuracy (phylum rank; among the samples the tool classified; SI Table S5; all distantly related test sequences) | Accuracy (phylum rank; among the samples the tool classified; SI Table S5; test sequences with at least one megablast hit to the training partition (removed by the filter)) | Accuracy (phylum rank; among the samples the tool classified; SI Table S5; test sequences without a megablast hit to the training partition) | Accuracy (superkingdom rank; among the samples the tool classified; SI Table S5; all distantly related test sequences) | Accuracy (superkingdom rank; among the samples the tool classified; SI Table S5; test sequences with at least one megablast hit to the training partition (removed by the filter)) | Accuracy (superkingdom rank; among the samples the tool classified; SI Table S5; test sequences without a megablast hit to the training partition) | Proportion (phylum rank; share of test samples the tool returned a prediction for) | Proportion (superkingdom rank; share of test samples the tool returned a prediction for) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| sourmash (Distantly related dataset, Mock et al. 2022)Evaluation | 91.7% (best in column) | 99.6% (best in column) | 3.48% | 25.1% | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | 0.16% | 0.17% |
| minimap2 (Distantly related dataset, Mock et al. 2022)Evaluation | 77.1% | 94.4% | 20% | 44.1% | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | 19.9% | 19.9% |
| MMseqs2 (Distantly related dataset, Mock et al. 2022)Evaluation | 75% | 90.7% | 41.4% | 62.8% | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | 52.8% | 52.8% |
| Kraken2 (Distantly related dataset, Mock et al. 2022)Evaluation | 68.3% | 90.7% | 19.5% | 44.4% | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | 21.1% | 21.6% |
| MMseqs2 + BERTax (Distantly related dataset, Mock et al. 2022)Evaluation | 66.6% | 89.4% | 60.3% (best in column) | 87.6% | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | 100% | 100% |
| BERTax (Distantly related dataset, Mock et al. 2022)Evaluation | 60.1% | 89% | 54.1% | 90.1% (best in column) | 60.1% (best in column) | 63.5% | 59.1% (best in column) | 89% (best in column) | 93.5% | 88.3% (best in column) | 100% | 100% |
| MMseqs2 taxonomy* (Distantly related dataset, Mock et al. 2022)Evaluation | 59.2% | 80.5% | 43.4% | 67.5% | 59.2% | 80.4% (best in column) | 53.3% | 80.5% | 95.8% (best in column) | 74.5% | 71.4% | 71.4% |
| MMseqs2 + DeepMicrobes (Distantly related dataset, Mock et al. 2022)Evaluation | 56.3% | 78.2% | 50.5% | 76.9% | Not reported | Not reported | Not reported | Not reported | Not reported | Not reported | 100% | 100% |
| DeepMicrobes (Distantly related dataset, Mock et al. 2022)Evaluation | 42% | 68.4% | 36.6% | 67.3% | 42% | 47.8% | 39.9% | 68.4% | 81.1% | 63.2% | 100% | 100% |
All 72 result rows with coverage, uncertainty and sources
Filter evaluations
Applied filters: All linked evaluations
Source checking is not independent reproduction. Release 2026-10-22-1b7cf027e4f6.
Methods and evaluation design
Procedure, tasks and evaluated configurations
Recorded evaluations
Each evaluation records what was tested and under which conditions.
- BERTax on the BERTax distantly related dataset
- DeepMicrobes on the BERTax distantly related dataset
- Kraken2 on the BERTax distantly related dataset
- minimap2 on the BERTax distantly related dataset
- MMseqs2 + BERTax on the BERTax distantly related dataset
- MMseqs2 + DeepMicrobes on the BERTax distantly related dataset
- MMseqs2 on the BERTax distantly related dataset
- MMseqs2 taxonomy* on the BERTax distantly related dataset
- sourmash on the BERTax distantly related dataset
Baseline coverage
Reference methods help show what a model adds beyond simple controls. We track a null control and a conventional method for each protocol.
0 of 2 active baseline roles have published Rewire measurements in this release. Measurements on a selected protocol do not establish coverage of an entire suite.
No execution recipe linked to this protocol. Recipe availability does not establish a completed evaluation.
- Author-reported evaluations
- 9
Literature evidence is not a Rewire measurement. Executed but unpublished runs and private review status are not included.
Null control
Proposed control: requires review
Training-set class prior where supervised fitting is permitted
Protocol-specific applicability, permitted inputs, access, split, evaluator and execution requirements need review before implementation or execution.
This is a suggested selection rule, not a validated method or a measured score.
Conventional reference
Proposed control: requires review
Regularised classifier on simple permitted features, or protocol's conventional reference
Protocol-specific applicability, permitted inputs, access, split, evaluator and execution requirements need review before implementation or execution.
This is a suggested selection rule, not a validated method or a measured score.
Protocol coverage CSV (gzip) · Model evaluation matrix (gzip) · Source table (gzip) · Release and checksums (gzip)
Coverage is derived from release 2026-10-22-1b7cf027e4f6. Source citations describe the original records; they do not validate an unreviewed baseline proposal. No results have been generated by this audit.
Run instructions
No runnable recipe has been reviewed for this protocol. Dataset access, model requirements, licences and compute requirements must be checked against its sources before execution.
Strengths, limitations and unresolved questions
Evidence
Source checking verifies the cited claim or transcription. It does not establish independent reproduction.
Evidence table
Inspect claims, sources and review details
Trace each statement to its source and review. A context-only reference supports the record generally; it does not verify an individual field. Source checking does not reproduce an experiment.
One row per statement and cited source. Multiple citations are not independent evaluations. Shared locators are labelled explicitly.
0 evidence rows matching the loaded filters
| Property and statement | Original source and location | Review and provenance |
|---|
No evidence rows match these filters. Choose another scope or clear the search.
Sources and history
Release 2026-10-22-1b7cf027e4f6 · Record review: source checked
3 source records and release history
- Taxonomic classification of DNA sequences beyond sequence similarity using deep neural networks · Original source · PNAS 119(35):e2122636119, published 2022-08-26; PMC9436379 full-text XML
- Taxonomic classification of DNA sequences beyond sequence similarity using deep neural networks (PNAS PDF) · Original source · PNAS 119(35):e2122636119 version-of-record PDF (10 pages)
- BERTax SI Appendix (pnas.2122636119.sapp.pdf) · Original source · SI Appendix as distributed with PMC9436379 (member pnas.2122636119.sapp.pdf of the Europe PMC supplementaryFiles ZIP)
Technical metadata and extraction receipts
Stable ID: dna-pathogen-20261012-bertax-protocol-distantly
- areas
- microbes-communities
- contexts
- research
- protocol
- Only each tool's top prediction per sample is used. For the PR curves and AveP, samples a tool does not classify are treated as random guesses. Macro AveP is the mean of per-class average precision, one-versus-all.
- version
- Table 1; SI Table S2; SI Table S5
- metric definition
- Macro AveP: unweighted mean over classes of per-class average precision of the PR curve. acc.: accuracy among the samples a tool classified; prop.: proportion of samples a tool returned a prediction for.
- source locator
- SI Appendix 'Details to PR-curve plots' (General Remarks; macro-average PR curves); Table 1 footnote; SI Table S2 caption
- limitations
- SI Tables S2 and S5 do not define the denominator of 'acc.'. It is read as the samples a tool classified: sourmash prints 99.56 superkingdom accuracy with 0.17% of the distantly related set classified, which is impossible over all samples, and the Table S5 caption notes that MMseqs2 taxonomy* predicts only 71.37% of sequences. Read 'acc.' together with 'prop.'.; Macro AveP counts samples a tool leaves unclassified as random guesses. BERTax and DeepMicrobes classify every sample and the database tools fewer, so AveP partly reflects how many samples each tool classifies.
- review
- method: source-hash-verification; scope-review; reviewer: claude; reviewer note: Separate Claude review agent (rev-mlvb), independent of the extracting agent; no human review claimed; date: 2026-10-12; artifact sha256: c0bb2f1bba8680b300076292389b39346ae58bea4bdf28647dd9462333b3bcd7; retrieval url: https://www.ebi.ac.uk/europepmc/webservices/rest/PMC9436379/fullTextXML; note: Protocol description checked against the article XML and SI; metric definition and limitations corrected by the reviewer.
Related records
- uses data: BERTax Distantly related dataset (Mock et al. 2022)
- part of: BERTax taxonomic classification of genome fragments (Mock et al. 2022)
- assessment: BERTax on the BERTax distantly related dataset
- assessment: DeepMicrobes on the BERTax distantly related dataset
- assessment: Kraken2 on the BERTax distantly related dataset
- assessment: minimap2 on the BERTax distantly related dataset
- assessment: MMseqs2 + BERTax on the BERTax distantly related dataset
- assessment: MMseqs2 + DeepMicrobes on the BERTax distantly related dataset
- assessment: MMseqs2 on the BERTax distantly related dataset
- assessment: MMseqs2 taxonomy* on the BERTax distantly related dataset
- assessment: sourmash on the BERTax distantly related dataset
- assessed by: Select a DNA pathogen-identification workflow for diagnostic testing