borzoi-ensemble: variant scores + random forest (paper Table 4): Supervised distance-balanced eQTL causality
Train the source’s random-forest classifier on the same balanced causality dataset and score the held-out chromosome test set.
Evaluation procedure
Train the source’s random-forest classifier on the same balanced causality dataset and score the held-out chromosome test set.
- Model
- borzoi-ensemble: variant scores + random forest (paper Table 4)
- Benchmark
- Supervised distance-balanced eQTL causality (AlphaGenome paper)
- Dataset
- Supervised distance-balanced eQTL causality: evaluated data subset
- origin
- Author-reported evaluation
- configuration
- borzoi-ensemble: variant scores + random forest (paper Table 4)
- protocol id
- alphagenome-2026-t4-protocol-12
- dataset version
- Not reported
- split
- Supervised training chromosomes are 1,4,7,8,10,13,15; validation chromosomes 2,5,11,14,17,20,22,X; test chromosomes 3,6,9,12,16,18,19,21. This is downstream evaluation separation, not a held-out reference-genome claim for the distilled backbone. The task additionally describes five-fold random-forest training excluding test chromosomes, followed by mean test auROC across those fits; exact fold membership is not enumerated.
- population
- Reprocessed GTEx SuSiE eQTL-catalogue SNVs; positives PIP≥0.9, negatives PIP≤0.01, GENCODEv46 genes, controls downsampled within log-distance-to-TSS bins.
- inputs
- Not reported
- adaptation
- borzoi-ensemble: variant scores + random forest (paper Table 4)
- metric implementation
- Not reported
- aggregation
- auROC per tissue weighted by the tissue’s variant count; the supervised analysis additionally averages auROC across five fitted folds.
- budget
- Not reported
Metadata review: needs review. Unreported conditions prevent automatic comparisons.
Evaluation results
Release 2026-09-17-a757f4af4277 · 1 evaluation · 1 metric row. Different protocols are not a single leaderboard.
| Metric and finding | Coverage and uncertainty | Evidence |
|---|---|---|
| borzoi-ensemble: variant scores + random forest (paper Table 4): Supervised distance-balanced eQTL causality Model: borzoi-ensemble: variant scores + random forest (paper Table 4) · Benchmark: Supervised distance-balanced eQTL causality (AlphaGenome paper) · Dataset: Supervised distance-balanced eQTL causality: evaluated data subset Train the source’s random-forest classifier on the same balanced causality dataset and score the held-out chromosome test set. Author-reported evaluation · Evaluation metadata: needs review | ||
| 0.759224069094061 tissue_weighted_mean_auroc Unit: dimensionless · Direction: higher Aggregation: auROC per tissue weighted by the tissue’s variant count; the supervised analysis additionally averages auROC across five fitted folds. | Uncertainty: unreported Scored: Not reported · Eligible: Not reported | source checkedAlphaGenome Nature 2026 supplementary comparison tables · 'Suppl Table 4 Variant performan'!L14 Source checking is not independent reproduction. |
Evidence table
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.
63 evidence rows matching the loaded filters
| Property and statement | Original source and location | Review and provenance |
|---|---|---|
| attributes.comparison.adaptation borzoi-ensemble: variant scores + random forest (paper Table 4) Context-only references | alphagenome: Journal full-text XML No field-specific location recorded Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: Retrieved page snapshot; no immutable publisher revision supplied | not individually reviewed No individual claim review recorded author reported Audit detailsField: Source artifact SHA-256: Hash scope: SHA-256 of retrieved original artifact bytes Format: original_artifact |
| attributes.comparison.adaptation borzoi-ensemble: variant scores + random forest (paper Table 4) Context-only references | AlphaGenome Nature 2026 supplementary methods No field-specific location recorded Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: Supplement to Nature version of record, 28 January 2026; content hash pinned | not individually reviewed No individual claim review recorded author reported Audit detailsField: Source artifact SHA-256: Hash scope: SHA-256 of exact publisher PDF bytes Format: original_pdf |
| attributes.comparison.adaptation borzoi-ensemble: variant scores + random forest (paper Table 4) Context-only references | AlphaGenome Nature 2026 supplementary comparison tables No field-specific location recorded Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: Nature version of record, 28 January 2026 | not individually reviewed No individual claim review recorded author reported Audit detailsField: Source artifact SHA-256: Hash scope: SHA-256 of exact retrieved original artifact bytes Format: xlsx |
| attributes.comparison.aggregation auROC per tissue weighted by the tissue’s variant count; the supervised analysis additionally averages auROC across five fitted folds. Context-only references | alphagenome: Journal full-text XML No field-specific location recorded Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: Retrieved page snapshot; no immutable publisher revision supplied | not individually reviewed No individual claim review recorded author reported Audit detailsField: Source artifact SHA-256: Hash scope: SHA-256 of retrieved original artifact bytes Format: original_artifact |
| attributes.comparison.aggregation auROC per tissue weighted by the tissue’s variant count; the supervised analysis additionally averages auROC across five fitted folds. Context-only references | AlphaGenome Nature 2026 supplementary methods No field-specific location recorded Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: Supplement to Nature version of record, 28 January 2026; content hash pinned | not individually reviewed No individual claim review recorded author reported Audit detailsField: Source artifact SHA-256: Hash scope: SHA-256 of exact publisher PDF bytes Format: original_pdf |
| attributes.comparison.aggregation auROC per tissue weighted by the tissue’s variant count; the supervised analysis additionally averages auROC across five fitted folds. Context-only references | AlphaGenome Nature 2026 supplementary comparison tables No field-specific location recorded Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: Nature version of record, 28 January 2026 | not individually reviewed No individual claim review recorded author reported Audit detailsField: Source artifact SHA-256: Hash scope: SHA-256 of exact retrieved original artifact bytes Format: xlsx |
| attributes.comparison.budget No value recorded Context-only references | alphagenome: Journal full-text XML No field-specific location recorded Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: Retrieved page snapshot; no immutable publisher revision supplied | missing or unspecified No individual claim review recorded author reported Audit detailsField: Source artifact SHA-256: Hash scope: SHA-256 of retrieved original artifact bytes Format: original_artifact |
| attributes.comparison.budget No value recorded Context-only references | AlphaGenome Nature 2026 supplementary methods No field-specific location recorded Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: Supplement to Nature version of record, 28 January 2026; content hash pinned | missing or unspecified No individual claim review recorded author reported Audit detailsField: Source artifact SHA-256: Hash scope: SHA-256 of exact publisher PDF bytes Format: original_pdf |
| attributes.comparison.budget No value recorded Context-only references | AlphaGenome Nature 2026 supplementary comparison tables No field-specific location recorded Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: Nature version of record, 28 January 2026 | missing or unspecified No individual claim review recorded author reported Audit detailsField: Source artifact SHA-256: Hash scope: SHA-256 of exact retrieved original artifact bytes Format: xlsx |
| attributes.comparison.dataset_version No value recorded Context-only references | alphagenome: Journal full-text XML No field-specific location recorded Shared locator for this statement’s cited sources; not a separate locator for each citation. Version: Retrieved page snapshot; no immutable publisher revision supplied | missing or unspecified No individual claim review recorded author reported Audit detailsField: Source artifact SHA-256: Hash scope: SHA-256 of retrieved original artifact bytes Format: original_artifact |
Sources and history
Release 2026-09-17-a757f4af4277 · Record review: needs review
- AlphaGenome Nature 2026 supplementary comparison tables · Original source · Nature version of record, 28 January 2026
- AlphaGenome Nature 2026 supplementary methods · Original source · Supplement to Nature version of record, 28 January 2026; content hash pinned
- alphagenome: Journal full-text XML · Original source · Retrieved page snapshot; no immutable publisher revision supplied
Technical metadata and extraction receipts
Stable ID: alphagenome-2026-evaluation-4e749d74191964fa
- areas
- dna-genomes
- origin
- author_reported
- protocol
- Train the source’s random-forest classifier on the same balanced causality dataset and score the held-out chromosome test set.
- version
- borzoi-ensemble: variant scores + random forest (paper Table 4)
- source evaluation index
- 12
- source table
- 4
- comparison
- protocol id: alphagenome-2026-t4-protocol-12; dataset version: Not reported; split: Supervised training chromosomes are 1,4,7,8,10,13,15; validation chromosomes 2,5,11,14,17,20,22,X; test chromosomes 3,6,9,12,16,18,19,21. This is downstream evaluation separation, not a held-out reference-genome claim for the distilled backbone. The task additionally describes five-fold random-forest training excluding test chromosomes, followed by mean test auROC across those fits; exact fold membership is not enumerated.; population: Reprocessed GTEx SuSiE eQTL-catalogue SNVs; positives PIP≥0.9, negatives PIP≤0.01, GENCODEv46 genes, controls downsampled within log-distance-to-TSS bins.; inputs: Not reported; adaptation: borzoi-ensemble: variant scores + random forest (paper Table 4); metric implementation: Not reported; aggregation: auROC per tissue weighted by the tissue’s variant count; the supervised analysis additionally averages auROC across five fitted folds.; budget: Not reported
- context
- allowed inputs: Absolute multimodal AlphaGenome variant features, with target-gene selection for gene-specific scores.; limitations: Fine-mapping labels represent putative causality. Distance balancing changes the sampled population and does not establish calibration to genome-wide prevalence. Do not equate the supervised feature pipeline with the RNA-only scorer.
- missing metadata
- dataset version: unextracted; exact input manifest: unextracted; inference budget: unreported; scoring failures: unreported
Related records
- model: borzoi-ensemble: variant scores + random forest (paper Table 4)
- benchmark: Supervised distance-balanced eQTL causality (AlphaGenome paper)
- dataset: Supervised distance-balanced eQTL causality: evaluated data subset
- evaluation: borzoi-ensemble: variant scores + random forest (paper Table 4): Supervised distance-balanced eQTL causality, tissue weighted mean auroc