scTab deep-ensemble uncertainty ROC-AUC: known-type vs. absent-type (Group 1 vs. Group 3)
Donor-held-out within the same pooled 249-dataset CELLxGENE corpus used for Table 1a; not a whole-study or whole-platform holdout. A 5-model deep ensemble is used: predicted probabilities are averaged across the 5 independently trained models first, and the uncertainty score is then computed as 1 minus the maximum of that averaged probability vector. This score is used to distinguish cells of known types correctly predicted by scTab (Group 1) from cells of types excluded entirely from training for being too rare (Group 3, described by the source as out-of-distribution/absent). This measures unknown/absent-type detection; it is a separate endpoint from known-type error detection (see ucc-research-protocol-sctab-uncertainty-error) and must not be merged with it.
Overview
Donor-held-out within the same pooled 249-dataset CELLxGENE corpus used for Table 1a; not a whole-study or whole-platform holdout. A 5-model deep ensemble is used: predicted probabilities are averaged across the 5 independently trained models first, and the uncertainty score is then computed as 1 minus the maximum of that averaged probability vector. This score is used to distinguish cells of known types correctly predicted by scTab (Group 1) from cells of types excluded entirely from training for being too rare (Group 3, described by the source as out-of-distribution/absent). This measures unknown/absent-type detection; it is a separate endpoint from known-type error detection (see ucc-research-protocol-sctab-uncertainty-error) and must not be merged with it.
Consult the linked sources for architecture or protocol details. Missing evidence is not evidence of a missing capability.
1 recorded evaluation, 1 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
1 evaluation · 1 result. Different protocols are not a single leaderboard.
Filter evaluations
Applied filters: All linked evaluations
| Tested configuration | Protocol and dataset | Finding | Evidence and details |
|---|---|---|---|
| Configuration: scTab — scTab Table 1 | Protocol: scTab deep-ensemble uncertainty ROC-AUC: known-type vs. absent-type (Group 1 vs. Group 3) Dataset: scTab processed CELLxGENE 2023-05-15 donor holdouts | 0.782 roc-auc fraction · higher Uncertainty: Not reported Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourcesctab uncertainty absent-type discrimination Not reported Aggregation: Not reported scTab: Scaling cross-tissue single-cell annotation models · Methods, "Uncertainty quantification for scTab model": "Group 1 and Group 3 can be separated with an ROC-AUC score of 0.782" |
Source checking is not independent reproduction. Release 2026-10-06-161b59a1d02c.
Methods and evaluation design
Procedure, tasks and evaluated configurations
Recorded evaluations
Each evaluation records what was tested and under which conditions.
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
- 1
Literature evidence is not a Rewire measurement. Executed but unpublished runs and private review status are not included.
Null control
Proposed control: requires review
Select a task-valid null control after reviewing inputs and metric
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
Select an upstream conventional reference after reviewing the full protocol
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 · Model evaluation matrix · Source table · Release and checksums
Coverage is derived from release 2026-10-06-161b59a1d02c. 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
View linked audit checks and correction history
Release 2026-10-06-161b59a1d02c · Record review: source checked
1 source records and release history
- scTab: Scaling cross-tissue single-cell annotation models · Original source · 10.1038/s41467-024-51059-5; published article XML retrieved 2026-09-30
Technical metadata and extraction receipts
Stable ID: ucc-research-protocol-sctab-uncertainty-absent
- review
- method: automated_source_review; actor: Claude Sonnet cell-type-annotation-transfer evidence-research worker; reviewed at: 2026-10-06T22:08:45Z; note: Source-backed primary-text transcription and independent re-verification against the already-catalogued scTab source (byte-identical SHA-256 re-check), continuing the bounded 2026-10-06 research dossier (docs/omics/evidence-research/cell-type-annotation-transfer-2026-10-06.md). Revised after an independent review found the averaging order, ensemble terminology and training-setting link needed correction; see below. No new model execution, independent experimental replication, or qualified human scientific review.
- source locator
- Methods, "Uncertainty quantification for scTab model"
- uncertainty mechanism
- Quoted verbatim, Methods, "Uncertainty quantification for scTab model": "one just averages the predicted probabilities across several networks that were independently trained (each with a different random initialization of the weights). In our case, we averaged the predictions across 5 models." The formula is given as 1 minus the maximum predicted probability. Order, exactly as stated: predicted probabilities are averaged across the 5 independently trained models first; the uncertainty score is then 1 minus the maximum of that averaged probability vector. This is not the same computation as averaging 5 separately computed per-model uncertainty scores. The source does not state temperature scaling for this computation; none is claimed here.
- limitations
- Donor-level holdout within one pooled 249-dataset corpus (10x-related assays only; rare types filtered); not a whole-study or whole-platform holdout. Whether individual datasets/studies contribute donors to more than one split is not directly confirmed in the retrieved text.; Single ROC-AUC value; no rejection-at-coverage curve and no demonstrated calibration of any decision threshold.; Exact number of cell types and cells comprising Group 3 (absent-type) is not stated in the retrieved main text; this is an explicit missing denominator, not invented.; This subsection is titled "Uncertainty quantification for scTab model" but does not independently restate the scTab model's training settings (e.g., data augmentation) for the ensemble members used in this specific analysis; the link to ucc-research-config-sctab-sctab reflects that subsection heading, not a repeated confirmation of training configuration here.; Reference/training labels are author annotations, not independent ground truth.; Automated source review only; independent human scientific review remains outstanding.