Source

Genomic heterogeneity inflates the performance of variant pathogenicity predictions (Lu, Liu, Lin and Brandes, bioRxiv 2025, v2)

bioRxiv preprint version 2, posted 2025-09-10 (version 1 posted 2025-09-08); full-text PDF (21 pages) · Preprint

DOI: 10.1101/2025.09.05.674459

What this source contributed

Records in this release that cite this source. A record can cite several sources.

1 method
21 configurations
1 protocol
1 dataset
15 dataset subsets
132 evaluations
132 results
3 evidence claims

Results

Values as the source printed them. Compare results only under compatible protocols, datasets and metrics.

Model or configuration	Metric	Value	Dataset	Evaluation
AlphaGenome all outputs, quantile, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign canonical splice SNVs (N=11,047); AlphaGenome scorer 'All outputs (quantile)')	0.614 unitless	Lu et al. 2025 ClinVar benchmark: splice variants scored by AlphaGenome (Supplementary Figure S2)	AlphaGenome All outputs (quantile) on ClinVar splice variants (Lu et al. 2025 Supplementary Figure S2)
AlphaGenome splice composite, quantile scores (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign canonical splice SNVs (N=11,047); AlphaGenome scorer 'All splice outputs (quantile)')	0.614 unitless	Lu et al. 2025 ClinVar benchmark: splice variants scored by AlphaGenome (Supplementary Figure S2)	AlphaGenome All splice outputs (quantile) on ClinVar splice variants (Lu et al. 2025 Supplementary Figure S2)
AlphaGenome splice composite, raw scores (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign canonical splice SNVs (N=11,047); AlphaGenome scorer 'All splice outputs (raw)')	0.708 unitless	Lu et al. 2025 ClinVar benchmark: splice variants scored by AlphaGenome (Supplementary Figure S2)	AlphaGenome All splice outputs (raw) on ClinVar splice variants (Lu et al. 2025 Supplementary Figure S2)
AlphaGenome all outputs, quantile, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, 3' UTR)	0.789 unitless	Lu et al. 2025 ClinVar benchmark: 3' UTR (Figure 1)	AlphaGenome on ClinVar 3' UTR SNVs (Lu et al. 2025 Figure 1)
AlphaGenome all outputs, quantile, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, 5' UTR)	0.580 unitless	Lu et al. 2025 ClinVar benchmark: 5' UTR (Figure 1)	AlphaGenome on ClinVar 5' UTR SNVs (Lu et al. 2025 Figure 1)
AlphaGenome all outputs, quantile, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, All variants)	0.310 unitless	Lu et al. 2025 ClinVar benchmark: All variants (Figure 1)	AlphaGenome on ClinVar All variants SNVs (Lu et al. 2025 Figure 1)
AlphaGenome all outputs, quantile, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Intron (Non-splice))	0.960 unitless	Lu et al. 2025 ClinVar benchmark: Intron (Non-splice) (Figure 1)	AlphaGenome on ClinVar Intron (Non-splice) SNVs (Lu et al. 2025 Figure 1)
AlphaGenome all outputs, quantile, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Missense & 3'UTR)	0.508 unitless	Lu et al. 2025 ClinVar benchmark: Missense & 3'UTR (Figure 1)	AlphaGenome on ClinVar Missense & 3'UTR SNVs (Lu et al. 2025 Figure 1)
AlphaGenome all outputs, quantile, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Missense & Intron (Non-splice))	0.583 unitless	Lu et al. 2025 ClinVar benchmark: Missense & Intron (Non-splice) (Figure 1)	AlphaGenome on ClinVar Missense & Intron (Non-splice) SNVs (Lu et al. 2025 Figure 1)
AlphaGenome all outputs, quantile, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Missense)	0.560 unitless	Lu et al. 2025 ClinVar benchmark: Missense (Figure 1)	AlphaGenome on ClinVar Missense SNVs (Lu et al. 2025 Figure 1)
AlphaGenome all outputs, quantile, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Noncoding)	0.991 unitless	Lu et al. 2025 ClinVar benchmark: Noncoding (Figure 1)	AlphaGenome on ClinVar Noncoding SNVs (Lu et al. 2025 Figure 1)
AlphaGenome all outputs, quantile, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, RNA gene)	0.851 unitless	Lu et al. 2025 ClinVar benchmark: RNA gene (Figure 1)	AlphaGenome on ClinVar RNA gene SNVs (Lu et al. 2025 Figure 1)
AlphaGenome all outputs, quantile, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Splice)	0.592 unitless	Lu et al. 2025 ClinVar benchmark: Splice (Figure 1)	AlphaGenome on ClinVar Splice SNVs (Lu et al. 2025 Figure 1)
AlphaGenome all outputs, quantile, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Start loss)	0.523 unitless	Lu et al. 2025 ClinVar benchmark: Start loss (Figure 1)	AlphaGenome on ClinVar Start loss SNVs (Lu et al. 2025 Figure 1)
AlphaGenome all outputs, quantile, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Stop gain)	0.389 unitless	Lu et al. 2025 ClinVar benchmark: Stop gain (Figure 1)	AlphaGenome on ClinVar Stop gain SNVs (Lu et al. 2025 Figure 1)
AlphaGenome all outputs, quantile, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Stop loss)	0.362 unitless	Lu et al. 2025 ClinVar benchmark: Stop loss (Figure 1)	AlphaGenome on ClinVar Stop loss SNVs (Lu et al. 2025 Figure 1)
AlphaGenome all outputs, quantile, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Synonymous)	0.919 unitless	Lu et al. 2025 ClinVar benchmark: Synonymous (Figure 1)	AlphaGenome on ClinVar Synonymous SNVs (Lu et al. 2025 Figure 1)
AlphaGenome splice junction, quantile, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign canonical splice SNVs (N=11,047); AlphaGenome scorer 'Splice junction (quantile)')	0.600 unitless	Lu et al. 2025 ClinVar benchmark: splice variants scored by AlphaGenome (Supplementary Figure S2)	AlphaGenome Splice junction (quantile) on ClinVar splice variants (Lu et al. 2025 Supplementary Figure S2)
AlphaGenome splice junction, raw, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign canonical splice SNVs (N=11,047); AlphaGenome scorer 'Splice junction (raw)')	0.669 unitless	Lu et al. 2025 ClinVar benchmark: splice variants scored by AlphaGenome (Supplementary Figure S2)	AlphaGenome Splice junction (raw) on ClinVar splice variants (Lu et al. 2025 Supplementary Figure S2)
AlphaGenome splice site, quantile, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign canonical splice SNVs (N=11,047); AlphaGenome scorer 'Splice site (quantile)')	0.683 unitless	Lu et al. 2025 ClinVar benchmark: splice variants scored by AlphaGenome (Supplementary Figure S2)	AlphaGenome Splice site (quantile) on ClinVar splice variants (Lu et al. 2025 Supplementary Figure S2)
AlphaGenome splice site, raw, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign canonical splice SNVs (N=11,047); AlphaGenome scorer 'Splice site (raw)')	0.706 unitless	Lu et al. 2025 ClinVar benchmark: splice variants scored by AlphaGenome (Supplementary Figure S2)	AlphaGenome Splice site (raw) on ClinVar splice variants (Lu et al. 2025 Supplementary Figure S2)
AlphaGenome splice site usage, quantile, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign canonical splice SNVs (N=11,047); AlphaGenome scorer 'Splice site usage (quantile)')	0.699 unitless	Lu et al. 2025 ClinVar benchmark: splice variants scored by AlphaGenome (Supplementary Figure S2)	AlphaGenome Splice site usage (quantile) on ClinVar splice variants (Lu et al. 2025 Supplementary Figure S2)
AlphaGenome splice site usage, raw, maximum (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign canonical splice SNVs (N=11,047); AlphaGenome scorer 'Splice site usage (raw)')	0.779 unitless	Lu et al. 2025 ClinVar benchmark: splice variants scored by AlphaGenome (Supplementary Figure S2)	AlphaGenome Splice site usage (raw) on ClinVar splice variants (Lu et al. 2025 Supplementary Figure S2)
AlphaMissense precomputed scores (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Missense & 3'UTR)	0.952 unitless	Lu et al. 2025 ClinVar benchmark: Missense & 3'UTR (Figure 1)	AlphaMissense on ClinVar Missense & 3'UTR SNVs (Lu et al. 2025 Figure 1)
AlphaMissense precomputed scores (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Missense & Intron (Non-splice))	0.924 unitless	Lu et al. 2025 ClinVar benchmark: Missense & Intron (Non-splice) (Figure 1)	AlphaMissense on ClinVar Missense & Intron (Non-splice) SNVs (Lu et al. 2025 Figure 1)
AlphaMissense precomputed scores (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Missense)	0.959 unitless	Lu et al. 2025 ClinVar benchmark: Missense (Figure 1)	AlphaMissense on ClinVar Missense SNVs (Lu et al. 2025 Figure 1)
AlphaMissense precomputed scores (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Stop gain)	0.849 unitless	Lu et al. 2025 ClinVar benchmark: Stop gain (Figure 1)	AlphaMissense on ClinVar Stop gain SNVs (Lu et al. 2025 Figure 1)
DNABERT-2 117M, CLS embedding L2 distance (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, 3' UTR)	0.499 unitless	Lu et al. 2025 ClinVar benchmark: 3' UTR (Figure 1)	DNABERT2 on ClinVar 3' UTR SNVs (Lu et al. 2025 Figure 1)
DNABERT-2 117M, CLS embedding L2 distance (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, 5' UTR)	0.414 unitless	Lu et al. 2025 ClinVar benchmark: 5' UTR (Figure 1)	DNABERT2 on ClinVar 5' UTR SNVs (Lu et al. 2025 Figure 1)
DNABERT-2 117M, CLS embedding L2 distance (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, All variants)	0.576 unitless	Lu et al. 2025 ClinVar benchmark: All variants (Figure 1)	DNABERT2 on ClinVar All variants SNVs (Lu et al. 2025 Figure 1)
DNABERT-2 117M, CLS embedding L2 distance (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Intron (Non-splice))	0.437 unitless	Lu et al. 2025 ClinVar benchmark: Intron (Non-splice) (Figure 1)	DNABERT2 on ClinVar Intron (Non-splice) SNVs (Lu et al. 2025 Figure 1)
DNABERT-2 117M, CLS embedding L2 distance (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Missense & 3'UTR)	0.546 unitless	Lu et al. 2025 ClinVar benchmark: Missense & 3'UTR (Figure 1)	DNABERT2 on ClinVar Missense & 3'UTR SNVs (Lu et al. 2025 Figure 1)
DNABERT-2 117M, CLS embedding L2 distance (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Missense & Intron (Non-splice))	0.449 unitless	Lu et al. 2025 ClinVar benchmark: Missense & Intron (Non-splice) (Figure 1)	DNABERT2 on ClinVar Missense & Intron (Non-splice) SNVs (Lu et al. 2025 Figure 1)
DNABERT-2 117M, CLS embedding L2 distance (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Missense)	0.484 unitless	Lu et al. 2025 ClinVar benchmark: Missense (Figure 1)	DNABERT2 on ClinVar Missense SNVs (Lu et al. 2025 Figure 1)
DNABERT-2 117M, CLS embedding L2 distance (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Noncoding)	0.433 unitless	Lu et al. 2025 ClinVar benchmark: Noncoding (Figure 1)	DNABERT2 on ClinVar Noncoding SNVs (Lu et al. 2025 Figure 1)
DNABERT-2 117M, CLS embedding L2 distance (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, RNA gene)	0.423 unitless	Lu et al. 2025 ClinVar benchmark: RNA gene (Figure 1)	DNABERT2 on ClinVar RNA gene SNVs (Lu et al. 2025 Figure 1)
DNABERT-2 117M, CLS embedding L2 distance (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Splice)	0.454 unitless	Lu et al. 2025 ClinVar benchmark: Splice (Figure 1)	DNABERT2 on ClinVar Splice SNVs (Lu et al. 2025 Figure 1)
DNABERT-2 117M, CLS embedding L2 distance (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Start loss)	0.389 unitless	Lu et al. 2025 ClinVar benchmark: Start loss (Figure 1)	DNABERT2 on ClinVar Start loss SNVs (Lu et al. 2025 Figure 1)
DNABERT-2 117M, CLS embedding L2 distance (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Stop gain)	0.483 unitless	Lu et al. 2025 ClinVar benchmark: Stop gain (Figure 1)	DNABERT2 on ClinVar Stop gain SNVs (Lu et al. 2025 Figure 1)
DNABERT-2 117M, CLS embedding L2 distance (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Stop loss)	0.570 unitless	Lu et al. 2025 ClinVar benchmark: Stop loss (Figure 1)	DNABERT2 on ClinVar Stop loss SNVs (Lu et al. 2025 Figure 1)
DNABERT-2 117M, CLS embedding L2 distance (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Synonymous)	0.552 unitless	Lu et al. 2025 ClinVar benchmark: Synonymous (Figure 1)	DNABERT2 on ClinVar Synonymous SNVs (Lu et al. 2025 Figure 1)
ESM1b 650M log-likelihood ratio, Brandes et al. pipeline (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Missense & 3'UTR)	0.906 unitless	Lu et al. 2025 ClinVar benchmark: Missense & 3'UTR (Figure 1)	ESM1b on ClinVar Missense & 3'UTR SNVs (Lu et al. 2025 Figure 1)
ESM1b 650M log-likelihood ratio, Brandes et al. pipeline (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Missense & Intron (Non-splice))	0.800 unitless	Lu et al. 2025 ClinVar benchmark: Missense & Intron (Non-splice) (Figure 1)	ESM1b on ClinVar Missense & Intron (Non-splice) SNVs (Lu et al. 2025 Figure 1)
ESM1b 650M log-likelihood ratio, Brandes et al. pipeline (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Missense)	0.919 unitless	Lu et al. 2025 ClinVar benchmark: Missense (Figure 1)	ESM1b on ClinVar Missense SNVs (Lu et al. 2025 Figure 1)
ESM1b 650M log-likelihood ratio, Brandes et al. pipeline (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Start loss)	0.681 unitless	Lu et al. 2025 ClinVar benchmark: Start loss (Figure 1)	ESM1b on ClinVar Start loss SNVs (Lu et al. 2025 Figure 1)
ESM1b 650M log-likelihood ratio, Brandes et al. pipeline (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Stop gain)	0.788 unitless	Lu et al. 2025 ClinVar benchmark: Stop gain (Figure 1)	ESM1b on ClinVar Stop gain SNVs (Lu et al. 2025 Figure 1)
ESM1v model 1 log-likelihood ratio, Brandes et al. pipeline (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Missense & 3'UTR)	0.870 unitless	Lu et al. 2025 ClinVar benchmark: Missense & 3'UTR (Figure 1)	ESM1v on ClinVar Missense & 3'UTR SNVs (Lu et al. 2025 Figure 1)
ESM1v model 1 log-likelihood ratio, Brandes et al. pipeline (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Missense & Intron (Non-splice))	0.769 unitless	Lu et al. 2025 ClinVar benchmark: Missense & Intron (Non-splice) (Figure 1)	ESM1v on ClinVar Missense & Intron (Non-splice) SNVs (Lu et al. 2025 Figure 1)
ESM1v model 1 log-likelihood ratio, Brandes et al. pipeline (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Missense)	0.886 unitless	Lu et al. 2025 ClinVar benchmark: Missense (Figure 1)	ESM1v on ClinVar Missense SNVs (Lu et al. 2025 Figure 1)
ESM1v model 1 log-likelihood ratio, Brandes et al. pipeline (Lu et al. 2025)	AUROC (ClinVar pathogenic vs benign SNVs, Start loss)	0.638 unitless	Lu et al. 2025 ClinVar benchmark: Start loss (Figure 1)	ESM1v on ClinVar Start loss SNVs (Lu et al. 2025 Figure 1)

Claims

Statements this source supports, each with where in the source it was checked.

AlphaGenome results depend on the scoring method: Evaluation finding Lu et al. report that AlphaGenome is highly sensitive to the method used to extract scores: with the recommended all-output maximum-quantile score it ranks first on non-splice intronic (0.960) and synonymous (0.919) ClinVar SNVs but reaches 0.592 on canonical splice SNVs, where the raw splice site usage score reaches 0.779 and the original publication's splice scorer 0.708 (Supplementary Figure S2). (Discussion paragraph 1 (page 10); Methods 'AlphaGenome' (pages 16 to 17); Figure 1; Supplementary Figure S2)
AlphaGenome scoring setup in Lu et al. 2025: Scoring setup Official AlphaGenome API (deepmind.google.com/science/alphagenome), 1 Mbp input context centred on the variant, organism human, all output modalities enabled (5,930 human genomic annotations across 11 modalities). Each output converted to quantiles; the variant score is the maximum percentile across all outputs. (Methods 'AlphaGenome' (pages 16 to 17))
Lu et al. 2025 values read from figure labels: Extraction method Figure 1 and Supplementary Figure S2 are the only places most values are printed. Figure 1 is vector art without a text layer and S2 is a raster image; every bar carries a printed numeric label and a named row, so the labels were transcribed from renders. Figure 2 (Evo2 score distributions) has no printed values and is not extracted. (Figure 1 (page 8); Figure 2 (page 9); Supplementary Figure S2 (page 21))

Models, benchmarks and other records

Methods (1)

Variant-type pathogenicity prior

Configurations (21)

Protocols (1)

ClinVar pathogenic versus benign SNVs within variant type, AUROC (Lu et al. 2025)

Datasets (1)

Lu et al. 2025 ClinVar SNV variant-type benchmark (259,600 SNVs)

Dataset subsets (15)

Evaluations (132)

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.

20 evidence rows matching the loaded filters

Claims, original sources and review scope · Release 2026-11-02-a7f03c9e55a5
Property and statement	Original source and location	Review and provenance
Artifact SHA-256 5b8e939257000811a6eaddd90d9de3e03bf1ac0e9015a38769df887ee01f9bf2 Source metadata	Genomic heterogeneity inflates the performance of variant pathogenicity predictions (Lu, Liu, Lin and Brandes, bioRxiv 2025, v2) Original source ↗ No field-specific location recorded Version: bioRxiv preprint version 2, posted 2025-09-10 (version 1 posted 2025-09-08); full-text PDF (21 pages) Retrieved: 2026-10-19T12:23:08Z	catalogued No individual claim review recorded Audit details Field: `attributes.artifact_sha256` Source artifact SHA-256: `5b8e939257000811a6eaddd90d9de3e03bf1ac0e9015a38769df887ee01f9bf2` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF. bioRxiv stamps the PDF on each download: a second retrieval at 2026-10-19T12:31:50Z had the same length (1,100,277 bytes), SHA-256 1a1aa1135ad8aa59bdb949bf8df75262eb6feba51b47f49439dbb819c426fbdd, differed in 32 bytes (ModDate) and had an identical pdftotext -layout text layer. Inspected artifact
Artifact url https://www.biorxiv.org/content/10.1101/2025.09.05.674459v2.full.pdf Source metadata	Genomic heterogeneity inflates the performance of variant pathogenicity predictions (Lu, Liu, Lin and Brandes, bioRxiv 2025, v2) Original source ↗ No field-specific location recorded Version: bioRxiv preprint version 2, posted 2025-09-10 (version 1 posted 2025-09-08); full-text PDF (21 pages) Retrieved: 2026-10-19T12:23:08Z	catalogued No individual claim review recorded Audit details Field: `attributes.artifact_url` Source artifact SHA-256: `5b8e939257000811a6eaddd90d9de3e03bf1ac0e9015a38769df887ee01f9bf2` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF. bioRxiv stamps the PDF on each download: a second retrieval at 2026-10-19T12:31:50Z had the same length (1,100,277 bytes), SHA-256 1a1aa1135ad8aa59bdb949bf8df75262eb6feba51b47f49439dbb819c426fbdd, differed in 32 bytes (ModDate) and had an identical pdftotext -layout text layer. Inspected artifact
Doi 10.1101/2025.09.05.674459 Source metadata	Genomic heterogeneity inflates the performance of variant pathogenicity predictions (Lu, Liu, Lin and Brandes, bioRxiv 2025, v2) Original source ↗ No field-specific location recorded Version: bioRxiv preprint version 2, posted 2025-09-10 (version 1 posted 2025-09-08); full-text PDF (21 pages) Retrieved: 2026-10-19T12:23:08Z	catalogued No individual claim review recorded Audit details Field: `attributes.doi` Source artifact SHA-256: `5b8e939257000811a6eaddd90d9de3e03bf1ac0e9015a38769df887ee01f9bf2` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF. bioRxiv stamps the PDF on each download: a second retrieval at 2026-10-19T12:31:50Z had the same length (1,100,277 bytes), SHA-256 1a1aa1135ad8aa59bdb949bf8df75262eb6feba51b47f49439dbb819c426fbdd, differed in 32 bytes (ModDate) and had an identical pdftotext -layout text layer. Inspected artifact
Hash scope SHA-256 of one retrieval of the bioRxiv PDF. bioRxiv stamps the PDF on each download: a second retrieval at 2026-10-19T12:31:50Z had the same length (1,100,277 bytes), SHA-256 1a1aa1135ad8aa59bdb949bf8df75262eb6feba51b47f49439dbb819c426fbdd, differed in 32 bytes (ModDate) and had an identical pdftotext -layout text layer. Source metadata	Genomic heterogeneity inflates the performance of variant pathogenicity predictions (Lu, Liu, Lin and Brandes, bioRxiv 2025, v2) Original source ↗ No field-specific location recorded Version: bioRxiv preprint version 2, posted 2025-09-10 (version 1 posted 2025-09-08); full-text PDF (21 pages) Retrieved: 2026-10-19T12:23:08Z	catalogued No individual claim review recorded Audit details Field: `attributes.hash_scope` Source artifact SHA-256: `5b8e939257000811a6eaddd90d9de3e03bf1ac0e9015a38769df887ee01f9bf2` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF. bioRxiv stamps the PDF on each download: a second retrieval at 2026-10-19T12:31:50Z had the same length (1,100,277 bytes), SHA-256 1a1aa1135ad8aa59bdb949bf8df75262eb6feba51b47f49439dbb819c426fbdd, differed in 32 bytes (ModDate) and had an identical pdftotext -layout text layer. Inspected artifact
Licence All rights reserved; no reuse allowed without permission (bioRxiv notice) Source metadata	Genomic heterogeneity inflates the performance of variant pathogenicity predictions (Lu, Liu, Lin and Brandes, bioRxiv 2025, v2) Original source ↗ No field-specific location recorded Version: bioRxiv preprint version 2, posted 2025-09-10 (version 1 posted 2025-09-08); full-text PDF (21 pages) Retrieved: 2026-10-19T12:23:08Z	catalogued No individual claim review recorded Audit details Field: `attributes.licence` Source artifact SHA-256: `5b8e939257000811a6eaddd90d9de3e03bf1ac0e9015a38769df887ee01f9bf2` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF. bioRxiv stamps the PDF on each download: a second retrieval at 2026-10-19T12:31:50Z had the same length (1,100,277 bytes), SHA-256 1a1aa1135ad8aa59bdb949bf8df75262eb6feba51b47f49439dbb819c426fbdd, differed in 32 bytes (ModDate) and had an identical pdftotext -layout text layer. Inspected artifact
Limitations 5 values Preprint, not peer reviewed. The authors' repository README (Brandes-Lab/VEP-eval, read 2026-10-19, not pinned) names a 2026 Nature Communications version with a rebuilt benchmark (ClinVar February 2026, MANE 1.5) and more models; its DOI did not resolve on 2026-10-19 and it was not extracted. Figure 1 and Supplementary Figure S2 have no text layer; their printed labels were transcribed from renders. No uncertainty is printed for any AUROC. Independent of the AlphaGenome, Evo 2, GPN-MSA and AlphaMissense developers. N. Brandes is an author of the ESM1b variant-effect pipeline reused here for the ESM models. Supplementary Table S1 (per-variant scores) was not retrieved. Source metadata	Genomic heterogeneity inflates the performance of variant pathogenicity predictions (Lu, Liu, Lin and Brandes, bioRxiv 2025, v2) Original source ↗ No field-specific location recorded Version: bioRxiv preprint version 2, posted 2025-09-10 (version 1 posted 2025-09-08); full-text PDF (21 pages) Retrieved: 2026-10-19T12:23:08Z	catalogued No individual claim review recorded Audit details Field: `attributes.limitations` Source artifact SHA-256: `5b8e939257000811a6eaddd90d9de3e03bf1ac0e9015a38769df887ee01f9bf2` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF. bioRxiv stamps the PDF on each download: a second retrieval at 2026-10-19T12:31:50Z had the same length (1,100,277 bytes), SHA-256 1a1aa1135ad8aa59bdb949bf8df75262eb6feba51b47f49439dbb819c426fbdd, differed in 32 bytes (ModDate) and had an identical pdftotext -layout text layer. Inspected artifact
Media type application/pdf Source metadata	Genomic heterogeneity inflates the performance of variant pathogenicity predictions (Lu, Liu, Lin and Brandes, bioRxiv 2025, v2) Original source ↗ No field-specific location recorded Version: bioRxiv preprint version 2, posted 2025-09-10 (version 1 posted 2025-09-08); full-text PDF (21 pages) Retrieved: 2026-10-19T12:23:08Z	catalogued No individual claim review recorded Audit details Field: `attributes.media_type` Source artifact SHA-256: `5b8e939257000811a6eaddd90d9de3e03bf1ac0e9015a38769df887ee01f9bf2` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF. bioRxiv stamps the PDF on each download: a second retrieval at 2026-10-19T12:31:50Z had the same length (1,100,277 bytes), SHA-256 1a1aa1135ad8aa59bdb949bf8df75262eb6feba51b47f49439dbb819c426fbdd, differed in 32 bytes (ModDate) and had an identical pdftotext -layout text layer. Inspected artifact
Publication status preprint Source metadata	Genomic heterogeneity inflates the performance of variant pathogenicity predictions (Lu, Liu, Lin and Brandes, bioRxiv 2025, v2) Original source ↗ No field-specific location recorded Version: bioRxiv preprint version 2, posted 2025-09-10 (version 1 posted 2025-09-08); full-text PDF (21 pages) Retrieved: 2026-10-19T12:23:08Z	catalogued No individual claim review recorded Audit details Field: `attributes.publication_status` Source artifact SHA-256: `5b8e939257000811a6eaddd90d9de3e03bf1ac0e9015a38769df887ee01f9bf2` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF. bioRxiv stamps the PDF on each download: a second retrieval at 2026-10-19T12:31:50Z had the same length (1,100,277 bytes), SHA-256 1a1aa1135ad8aa59bdb949bf8df75262eb6feba51b47f49439dbb819c426fbdd, differed in 32 bytes (ModDate) and had an identical pdftotext -layout text layer. Inspected artifact
Retrieved at 2026-10-19T12:23:08Z Source metadata	Genomic heterogeneity inflates the performance of variant pathogenicity predictions (Lu, Liu, Lin and Brandes, bioRxiv 2025, v2) Original source ↗ No field-specific location recorded Version: bioRxiv preprint version 2, posted 2025-09-10 (version 1 posted 2025-09-08); full-text PDF (21 pages) Retrieved: 2026-10-19T12:23:08Z	catalogued No individual claim review recorded Audit details Field: `attributes.retrieved_at` Source artifact SHA-256: `5b8e939257000811a6eaddd90d9de3e03bf1ac0e9015a38769df887ee01f9bf2` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF. bioRxiv stamps the PDF on each download: a second retrieval at 2026-10-19T12:31:50Z had the same length (1,100,277 bytes), SHA-256 1a1aa1135ad8aa59bdb949bf8df75262eb6feba51b47f49439dbb819c426fbdd, differed in 32 bytes (ModDate) and had an identical pdftotext -layout text layer. Inspected artifact
Source location Results; Figure 1 (page 8); Methods; Supplementary Figures S1 and S2 (pages 20 to 21) Source metadata	Genomic heterogeneity inflates the performance of variant pathogenicity predictions (Lu, Liu, Lin and Brandes, bioRxiv 2025, v2) Original source ↗ No field-specific location recorded Version: bioRxiv preprint version 2, posted 2025-09-10 (version 1 posted 2025-09-08); full-text PDF (21 pages) Retrieved: 2026-10-19T12:23:08Z	catalogued No individual claim review recorded Audit details Field: `attributes.source_locator` Source artifact SHA-256: `5b8e939257000811a6eaddd90d9de3e03bf1ac0e9015a38769df887ee01f9bf2` Hash scope: SHA-256 of one retrieval of the bioRxiv PDF. bioRxiv stamps the PDF on each download: a second retrieval at 2026-10-19T12:31:50Z had the same length (1,100,277 bytes), SHA-256 1a1aa1135ad8aa59bdb949bf8df75262eb6feba51b47f49439dbb819c426fbdd, differed in 32 bytes (ModDate) and had an identical pdftotext -layout text layer. Inspected artifact

How to read the evidence

Sources and history

Release 2026-11-02-a7f03c9e55a5 · Record review: source checked

Primary source retrieved and hashed for the 2026-10-19 AlphaGenome variant evidence extraction.

Release history

No supporting source is linked yet.

Read original source

Download this release (gzip)

Technical metadata and extraction receipts

Stable ID: lu-hetero-20261019-source-preprint-v2

areas: dna-genomes
contexts: clinical_research
url: https://doi.org/10.1101/2025.09.05.674459
artifact url: https://www.biorxiv.org/content/10.1101/2025.09.05.674459v2.full.pdf
doi: 10.1101/2025.09.05.674459
version: bioRxiv preprint version 2, posted 2025-09-10 (version 1 posted 2025-09-08); full-text PDF (21 pages)
retrieved at: 2026-10-19T12:23:08Z
artifact sha256: 5b8e939257000811a6eaddd90d9de3e03bf1ac0e9015a38769df887ee01f9bf2
publication status: preprint
media type: application/pdf
source locator: Results; Figure 1 (page 8); Methods; Supplementary Figures S1 and S2 (pages 20 to 21)
hash scope: SHA-256 of one retrieval of the bioRxiv PDF. bioRxiv stamps the PDF on each download: a second retrieval at 2026-10-19T12:31:50Z had the same length (1,100,277 bytes), SHA-256 1a1aa1135ad8aa59bdb949bf8df75262eb6feba51b47f49439dbb819c426fbdd, differed in 32 bytes (ModDate) and had an identical pdftotext -layout text layer.
limitations: Preprint, not peer reviewed. The authors' repository README (Brandes-Lab/VEP-eval, read 2026-10-19, not pinned) names a 2026 Nature Communications version with a rebuilt benchmark (ClinVar February 2026, MANE 1.5) and more models; its DOI did not resolve on 2026-10-19 and it was not extracted.; Figure 1 and Supplementary Figure S2 have no text layer; their printed labels were transcribed from renders.; No uncertainty is printed for any AUROC.; Independent of the AlphaGenome, Evo 2, GPN-MSA and AlphaMissense developers. N. Brandes is an author of the ESM1b variant-effect pipeline reused here for the ESM models.; Supplementary Table S1 (per-variant scores) was not retrieved.

Related records

subject: Lu et al. 2025 values read from figure labels

Suggest a correction