Protocol

EVEE ClinVar indel pathogenicity AUROC (Pearce et al. 2026 Figure 1d)

Evidence concern: excluded from comparisons

The column labelled 'NTv3' in Figure 1b and 1c is not the method the Methods describe. The manuscript repository's plotting code (scripts/figure1/_heatmap_common.py at e5e4a43) renames 'NTv3 cosine dist' to 'NTv3' and drops 'NTv3 subref probe', whereas Methods 'Baseline methods' describe NTv3 650M layer-11 embeddings with L2-regularised logistic regression under 5-fold gene-holdout cross-validation. Figure 1d's NTv3 column is the supervised subref probe. The SNV and deconfounded NTv3 values are recorded as a zero-shot embedding distance.Interpretable variant effect prediction from genomic foundation model representations (EVEE), bioRxiv v4 · Figure 1b and 1c, column 'NTv3' (page 3); Methods 'Baseline methods', NTv3 (page 8); repository scripts/figure1/_heatmap_common.py RENAME_MAP and DROP_METHODS

AlphaMissense's 'Overall' cell in Figure 1b and 1c is its missense AUROC. The plotting code blanks AlphaMissense outside missense and copies the Missense value into the Overall row (scripts/figure1/_heatmap_common.py, am_ov_mask), so it is computed on a different, missense-only population from the other methods' Overall cells.Interpretable variant effect prediction from genomic foundation model representations (EVEE), bioRxiv v4 · Figure 1b and 1c, row 'Overall', column 'AlphaMissense' (page 3); repository scripts/figure1/_heatmap_common.py prepare()

AUROC for pathogenic versus benign on 73,961 ClinVar indels, all strata pooled (Overall row).

4 evaluations · 4 results

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.

4 recorded evaluations, 4 metric rows. A comparison chart has not yet been validated for these results. The table retains the individual findings and their sources.

View coverage and remaining gaps across all benchmarks

Results

Results are available, but no reviewed comparison panel is linked in this release.

All evaluations

4 evaluations · 4 results. Different protocols are not a single leaderboard.

Protocol: EVEE ClinVar indel pathogenicity AUROC (Pearce et al. 2026 Figure 1d) · Dataset: EVEE ClinVar indel benchmark (73,961 indels)

Sorted by AUROC (ClinVar indels, all strata (Overall row); supervised on indels) (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	AUROC (ClinVar indels, all strata (Overall row); supervised on indels)	AUROC (ClinVar indels, all strata (Overall row); pre-computed CADD v1.7 InDel scores)	AUROC (ClinVar indels, all strata (Overall row); zero-shot transfer from SNV training)
Evo 2 7B mean-pooled embedding probe (Pearce et al. 2026)Evaluation	0.984 (best in column)	Not reported	Not reported
NTv3 650M layer-11 embedding probe (Pearce et al. 2026 Figure 1d)Evaluation	0.828	Not reported	Not reported
CADD v1.7 PHRED, pre-computed (Pearce et al. 2026)Evaluation	Not reported	0.98	Not reported
EVEE covariance probe on Evo 2 7B layer 27 (Pearce et al. 2026)Evaluation	Not reported	Not reported	0.991

All 4 result rows with coverage, uncertainty and sources

Filter evaluations

Tested model or methodProtocolDatasetEvaluation setupMetricEvidence origin

Applied filters: All linked evaluations

Exact evaluated configurations and original reported results
Tested configuration	Protocol and dataset	Finding	Evidence and details
Configuration: CADD v1.7 PHRED, pre-computed (Pearce et al. 2026)	Protocol: EVEE ClinVar indel pathogenicity AUROC (Pearce et al. 2026 Figure 1d) Dataset: EVEE ClinVar indel benchmark (73,961 indels)	0.98 AUROC (ClinVar indels, all strata (Overall row); pre-computed CADD v1.7 InDel scores) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The column labelled 'NTv3' in Figure 1b and 1c is not the method the Methods describe. The manuscript repository's plotting code (scripts/figure1/_heatmap_common.py at e5e4a43) renames 'NTv3 cosine dist' to 'NTv3' and drops 'NTv3 subref probe', whereas Methods 'Baseline methods' describe NTv3 650M layer-11 embeddings with L2-regularised logistic regression under 5-fold gene-holdout cross-validation. Figure 1d's NTv3 column is the supervised subref probe. The SNV and deconfounded NTv3 values are recorded as a zero-shot embedding distance.Interpretable variant effect prediction from genomic foundation model representations (EVEE), bioRxiv v4 · Figure 1b and 1c, column 'NTv3' (page 3); Methods 'Baseline methods', NTv3 (page 8); repository scripts/figure1/_heatmap_common.py RENAME_MAP and DROP_METHODS AlphaMissense's 'Overall' cell in Figure 1b and 1c is its missense AUROC. The plotting code blanks AlphaMissense outside missense and copies the Missense value into the Overall row (scripts/figure1/_heatmap_common.py, am_ov_mask), so it is computed on a different, missense-only population from the other methods' Overall cells.Interpretable variant effect prediction from genomic foundation model representations (EVEE), bioRxiv v4 · Figure 1b and 1c, row 'Overall', column 'AlphaMissense' (page 3); repository scripts/figure1/_heatmap_common.py prepare() Methods, coverage and source CADD v1.7 on the EVEE ClinVar indel benchmark (Pearce et al. 2026) gfm-guide-20261014-protocol-evee-clinvar-indel-auroc Aggregation: Not reported Interpretable variant effect prediction from genomic foundation model representations (EVEE), bioRxiv v4; EVEE manuscript repository: figures/figure1/fig1d_indel_heatmap.pdf; EVEE manuscript repository: scripts/figure1/fig1d_indel_heatmap.py · Figure 1d (page 3, raster; matched to repository figures/figure1/fig1d_indel_heatmap.pdf text layer), row 'Overall (n=73,961, 65.1%P)', column 'CADD v1.7'
Configuration: EVEE covariance probe on Evo 2 7B layer 27 (Pearce et al. 2026)	Protocol: EVEE ClinVar indel pathogenicity AUROC (Pearce et al. 2026 Figure 1d) Dataset: EVEE ClinVar indel benchmark (73,961 indels)	0.991 AUROC (ClinVar indels, all strata (Overall row); zero-shot transfer from SNV training) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The column labelled 'NTv3' in Figure 1b and 1c is not the method the Methods describe. The manuscript repository's plotting code (scripts/figure1/_heatmap_common.py at e5e4a43) renames 'NTv3 cosine dist' to 'NTv3' and drops 'NTv3 subref probe', whereas Methods 'Baseline methods' describe NTv3 650M layer-11 embeddings with L2-regularised logistic regression under 5-fold gene-holdout cross-validation. Figure 1d's NTv3 column is the supervised subref probe. The SNV and deconfounded NTv3 values are recorded as a zero-shot embedding distance.Interpretable variant effect prediction from genomic foundation model representations (EVEE), bioRxiv v4 · Figure 1b and 1c, column 'NTv3' (page 3); Methods 'Baseline methods', NTv3 (page 8); repository scripts/figure1/_heatmap_common.py RENAME_MAP and DROP_METHODS AlphaMissense's 'Overall' cell in Figure 1b and 1c is its missense AUROC. The plotting code blanks AlphaMissense outside missense and copies the Missense value into the Overall row (scripts/figure1/_heatmap_common.py, am_ov_mask), so it is computed on a different, missense-only population from the other methods' Overall cells.Interpretable variant effect prediction from genomic foundation model representations (EVEE), bioRxiv v4 · Figure 1b and 1c, row 'Overall', column 'AlphaMissense' (page 3); repository scripts/figure1/_heatmap_common.py prepare() Methods, coverage and source Evo2 Covariance on the EVEE ClinVar indel benchmark (Pearce et al. 2026) gfm-guide-20261014-protocol-evee-clinvar-indel-auroc Aggregation: Not reported Interpretable variant effect prediction from genomic foundation model representations (EVEE), bioRxiv v4; EVEE manuscript repository: figures/figure1/fig1d_indel_heatmap.pdf; EVEE manuscript repository: scripts/figure1/fig1d_indel_heatmap.py · Figure 1d (page 3, raster; matched to repository figures/figure1/fig1d_indel_heatmap.pdf text layer), row 'Overall (n=73,961, 65.1%P)', column 'Evo2 Covariance'
Configuration: Evo 2 7B mean-pooled embedding probe (Pearce et al. 2026)	Protocol: EVEE ClinVar indel pathogenicity AUROC (Pearce et al. 2026 Figure 1d) Dataset: EVEE ClinVar indel benchmark (73,961 indels)	0.984 AUROC (ClinVar indels, all strata (Overall row); supervised on indels) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The column labelled 'NTv3' in Figure 1b and 1c is not the method the Methods describe. The manuscript repository's plotting code (scripts/figure1/_heatmap_common.py at e5e4a43) renames 'NTv3 cosine dist' to 'NTv3' and drops 'NTv3 subref probe', whereas Methods 'Baseline methods' describe NTv3 650M layer-11 embeddings with L2-regularised logistic regression under 5-fold gene-holdout cross-validation. Figure 1d's NTv3 column is the supervised subref probe. The SNV and deconfounded NTv3 values are recorded as a zero-shot embedding distance.Interpretable variant effect prediction from genomic foundation model representations (EVEE), bioRxiv v4 · Figure 1b and 1c, column 'NTv3' (page 3); Methods 'Baseline methods', NTv3 (page 8); repository scripts/figure1/_heatmap_common.py RENAME_MAP and DROP_METHODS AlphaMissense's 'Overall' cell in Figure 1b and 1c is its missense AUROC. The plotting code blanks AlphaMissense outside missense and copies the Missense value into the Overall row (scripts/figure1/_heatmap_common.py, am_ov_mask), so it is computed on a different, missense-only population from the other methods' Overall cells.Interpretable variant effect prediction from genomic foundation model representations (EVEE), bioRxiv v4 · Figure 1b and 1c, row 'Overall', column 'AlphaMissense' (page 3); repository scripts/figure1/_heatmap_common.py prepare() Methods, coverage and source Evo2 Mean on the EVEE ClinVar indel benchmark (Pearce et al. 2026) gfm-guide-20261014-protocol-evee-clinvar-indel-auroc Aggregation: Not reported Interpretable variant effect prediction from genomic foundation model representations (EVEE), bioRxiv v4; EVEE manuscript repository: figures/figure1/fig1d_indel_heatmap.pdf; EVEE manuscript repository: scripts/figure1/fig1d_indel_heatmap.py · Figure 1d (page 3, raster; matched to repository figures/figure1/fig1d_indel_heatmap.pdf text layer), row 'Overall (n=73,961, 65.1%P)', column 'Evo2 Mean'
Configuration: NTv3 650M layer-11 embedding probe (Pearce et al. 2026 Figure 1d)	Protocol: EVEE ClinVar indel pathogenicity AUROC (Pearce et al. 2026 Figure 1d) Dataset: EVEE ClinVar indel benchmark (73,961 indels)	0.828 AUROC (ClinVar indels, all strata (Overall row); supervised on indels) unitless · higher Uncertainty: Not reported by the source Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons The column labelled 'NTv3' in Figure 1b and 1c is not the method the Methods describe. The manuscript repository's plotting code (scripts/figure1/_heatmap_common.py at e5e4a43) renames 'NTv3 cosine dist' to 'NTv3' and drops 'NTv3 subref probe', whereas Methods 'Baseline methods' describe NTv3 650M layer-11 embeddings with L2-regularised logistic regression under 5-fold gene-holdout cross-validation. Figure 1d's NTv3 column is the supervised subref probe. The SNV and deconfounded NTv3 values are recorded as a zero-shot embedding distance.Interpretable variant effect prediction from genomic foundation model representations (EVEE), bioRxiv v4 · Figure 1b and 1c, column 'NTv3' (page 3); Methods 'Baseline methods', NTv3 (page 8); repository scripts/figure1/_heatmap_common.py RENAME_MAP and DROP_METHODS AlphaMissense's 'Overall' cell in Figure 1b and 1c is its missense AUROC. The plotting code blanks AlphaMissense outside missense and copies the Missense value into the Overall row (scripts/figure1/_heatmap_common.py, am_ov_mask), so it is computed on a different, missense-only population from the other methods' Overall cells.Interpretable variant effect prediction from genomic foundation model representations (EVEE), bioRxiv v4 · Figure 1b and 1c, row 'Overall', column 'AlphaMissense' (page 3); repository scripts/figure1/_heatmap_common.py prepare() Methods, coverage and source NTv3 on the EVEE ClinVar indel benchmark (Pearce et al. 2026) gfm-guide-20261014-protocol-evee-clinvar-indel-auroc Aggregation: Not reported Interpretable variant effect prediction from genomic foundation model representations (EVEE), bioRxiv v4; EVEE manuscript repository: figures/figure1/fig1d_indel_heatmap.pdf; EVEE manuscript repository: scripts/figure1/fig1d_indel_heatmap.py · Figure 1d (page 3, raster; matched to repository figures/figure1/fig1d_indel_heatmap.pdf text layer), row 'Overall (n=73,961, 65.1%P)', column 'NTv3'

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.

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: 2
External evaluations: 2

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 (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

Claims, original sources and review scope · Release 2026-10-22-1b7cf027e4f6
Property and statement	Original source and location	Review and provenance

No evidence rows match these filters. Choose another scope or clear the search.

How to read the evidence

Sources and history

Release 2026-10-22-1b7cf027e4f6 · Record review: source checked

3 source records and release history

Interpretable variant effect prediction from genomic foundation model representations (EVEE), bioRxiv v4 · Original source · bioRxiv v4, posted 2026-07-11 (v1 posted 2026-04-11)
EVEE manuscript repository: figures/figure1/fig1d_indel_heatmap.pdf · Original source · goodfire-ai/evee-manuscript commit e5e4a43d60bb217819144c09ab509ed44f49cd5f
EVEE manuscript repository: scripts/figure1/fig1d_indel_heatmap.py · Original source · goodfire-ai/evee-manuscript commit e5e4a43d60bb217819144c09ab509ed44f49cd5f

Download this release (gzip)

Technical metadata and extraction receipts

Stable ID: gfm-guide-20261014-protocol-evee-clinvar-indel-auroc

areas: dna-genomes
contexts: research
protocol: Each method's score on the 73,961 ClinVar indels, AUROC for P/LP versus B/LB. The covariance probe was trained on SNVs only; Evo2 Mean and NTv3 are supervised on indels; CADD v1.7 InDel scores indels under 50 bp.
version: bioRxiv v4 Figure 1d
metric: auroc
limitations: Developer-reported: the authors built EVEE and three of them co-authored Evo 2.; Mixes zero-shot and supervised methods in one comparison (repository column names).; One value per method; no confidence interval is printed in Figure 1d.
source locator: Figure 1d (page 3); Methods 'Indel benchmark' (page 9); main text lines 141-143 (page 2)

Related records

uses data: EVEE ClinVar indel benchmark (73,961 indels)
assessment: CADD v1.7 on the EVEE ClinVar indel benchmark (Pearce et al. 2026)
assessment: Evo2 Covariance on the EVEE ClinVar indel benchmark (Pearce et al. 2026)
assessment: Evo2 Mean on the EVEE ClinVar indel benchmark (Pearce et al. 2026)
assessment: NTv3 on the EVEE ClinVar indel benchmark (Pearce et al. 2026)

Suggest a correction