rewire.it
benchmark · task

Held-out-clade classification

Can a sequence method generalize beyond taxa present during fitting?

0 evaluations · 0 metric rows

At a glance

Explanatory profile: limited source coverage · Automated source review, 2026-09-16. This does not change the review status of its results.

Data, procedure and scoring
PropertyDescription and evidence
Record typeGeneric task; proposed designProkBERT official resource; METAGENE-1 official resource; Evo 2 official resource; Kraken2 official resource · Catalogue task scope and proposed comparison design; concrete source protocol has not yet been extracted
InputsHold clades out of downstream fitting and reference databases; evaluate known ancestor labels or unknown-taxon detection, and audit pretraining overlap separately.ProkBERT official resource; METAGENE-1 official resource; Evo 2 official resource; Kraken2 official resource · Catalogue task scope and proposed comparison design; concrete source protocol has not yet been extracted
AssessmentNo single dataset, split or scoring implementation is fixed by this task recordProkBERT official resource; METAGENE-1 official resource; Evo 2 official resource; Kraken2 official resource · Catalogue task scope and proposed comparison design; concrete source protocol has not yet been extracted
DatasetsNot extracted or verified for this record.
OrganismsNot extracted or verified for this record.
AssaysNot extracted or verified for this record.
SplitsNot extracted or verified for this record.
AdaptationNot extracted or verified for this record.
BaselinesNot extracted or verified for this record.

How it works

Proposed evaluation design

Conceptual design only. This generic task has no fixed executable protocol and does not imply a completed run.

Proposed evaluation designChoose withheld clades. Then: Audit training and references. Then: Classify held-out sequences. Then: Assess ancestral or unknown labelsChoose withheld cladesAudit training and referencesClassify held-out sequencesAssess ancestral or unknownlabels
Read the diagram as text
  1. Choose withheld clades
  2. Audit training and references
  3. Classify held-out sequences
  4. Assess ancestral or unknown labels
ProkBERT official resource; METAGENE-1 official resource; Evo 2 official resource; Kraken2 official resource · Catalogue task scope and proposed comparison design; concrete source protocol has not yet been extracted

Proposed comparison design

Withhold the chosen clades from downstream training and reference databases, then separately audit overlap with model pretraining.

ProkBERT official resource; METAGENE-1 official resource; Evo 2 official resource; Kraken2 official resource · Catalogue task scope and proposed comparison design; concrete source protocol has not yet been extracted

Tested models and results

Release 2026-09-16-d74d282221a9 · 0 evaluations · 0 metric rows. Different protocols are not a single leaderboard.

No evaluations linked in this release.

Strengths and limitations

Profile review details

Catalogue extraction inspected; protocol claims remain limited to the cited evidence. Missing details are not presumed absent from the original paper.

Stable record: catalog-task-heldout-clade

Applicable tests and references

Applicability is distinct from a completed evaluation.

Sources and history

Release 2026-09-16-d74d282221a9 · Record review: discovered

Download this release
Technical metadata and extraction receipts

Stable ID: catalog-task-heldout-clade

areas
microbes-communities
entity level
task
version
Not reported
task
Held-out-clade classification
scope note
Hold clades out of downstream fitting and reference databases; evaluate known ancestor labels or unknown-taxon detection, and audit pretraining overlap separately.
missing metadata
protocol version: not_yet_extracted
Related records

Suggest a correction