Odds ratio (rare (singletons) versus common (MAF > 5%) gnomAD variants in the tail of deleterious scores, threshold at 30 false discoveries, one-sided Fisher's exact test; all variant types, rare subsampled to the common count; mean over three random seeds; zero-shot) of GPN-MSA ablation 'Window size = 64' (Supplementary Table 4) on gnomAD rare (singletons) vs common variants, ablation set (GPN-MSA paper)
37.5 Odds ratio (rare (singletons) versus common (MAF > 5%) gnomAD variants in the tail of deleterious scores, threshold at 30 false discoveries, one-sided Fisher's exact test; all variant types, rare subsampled to the common count; mean over three random seeds; zero-shot)
Methods
- Tested configuration
- GPN-MSA ablation 'Window size = 64' (Supplementary Table 4)
- Protocol
- gnomAD rare vs common enrichment in the deleterious tail, odds ratio (GPN-MSA Supplementary Table 4)
- Dataset
- gnomAD rare (singletons) vs common variants, ablation set (GPN-MSA paper)
- Procedure
- gnomAD rare vs common enrichment in the deleterious tail, odds ratio (GPN-MSA Supplementary Table 4)
- Evaluation
- Window size = 64 on gnomAD (GPN-MSA Supplementary Table 4)
- Coverage
- Not reported scored / Not reported eligible
- Uncertainty
- Not reported by the source: Only mean and max over seeds are printed
- Evidence
- Author-reported evaluation · source checkedGPN-MSA: an alignment-based DNA language model for genome-wide variant effect prediction (bioRxiv v2) · Supplementary Table 4 (PDF page 23), row 'Window size = 64', column 'gnomAD' / 'mean'
A source-checked result verifies the numerical transcription, not every model or protocol detail. Evaluation metadata: source checked. Source checked does not mean independently reproduced.
Reproduction
- Split
- Whole labelled set
- Adaptation
- zero-shot
- Scoring implementation
- Odds ratio from a one-sided Fisher's exact test on rare versus common variants above the 30-false-discovery threshold
No execution recipe has been verified for this exact configuration and evaluation. A benchmark's general instructions may use different inputs, splits or model settings.
Reproducing this published result requires matching its model configuration, data, split and scorer. Source checking or a successful smoke test does not establish score reproduction.
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.
1 evidence row matching the loaded filters
| Property and statement | Original source and location | Review and provenance |
|---|---|---|
| Reported result 37.478 Individual claims | GPN-MSA: an alignment-based DNA language model for genome-wide variant effect prediction (bioRxiv v2) Supplementary Table 4 (PDF page 23), row 'Window size = 64', column 'gnomAD' / 'mean' Version: bioRxiv v2, posted 2024-04-06 | source checked ["pdf-text-parse","independent-cell-check"] · 2026-10-14 author reported Audit detailsExtraction: Parsed the pdftotext 26.08.0 -layout text of a 2026-10-14 re-download of the v2 PDF (same length as the pinned file; SHA-256 c066dab122b359b90059a81f325fa2cd83122df14a700eae8d35021471c10978) with extract/build_gnomad.py; caption, column groups, 'mean max' header, the 17 row labels and the page asserted, and the row's ClinVar, COSMIC and OMIM cells asserted equal to the reviewed store values. Independent review 2026-10-14: matched. The 30-false-discovery threshold is from the Figure 2f caption, which the table caption says the ablation follows ('same approach'); recorded in the protocol limitations. Field: Source artifact SHA-256: Hash scope: SHA-256 of the PDF bytes as downloaded 2026-10-14T12:02:16Z. bioRxiv rewrites the XMP dates and document IDs, the Info ModDate and the trailer /ID on download, so a re-download does not reproduce the hash; the review found a fresh copy of the same length differing only in those 51 bytes, with an identical pdftotext -layout text layer. Extraction artifact SHA-256: |
Sources and history
Release 2026-10-22-1b7cf027e4f6 · Record review: source checked
1 source record and release history
- GPN-MSA: an alignment-based DNA language model for genome-wide variant effect prediction (bioRxiv v2) · Original source · bioRxiv v2, posted 2024-04-06
Technical metadata and extraction receipts
Stable ID: gfm-guides-20261014-gpnmsa-result-ablation-window-64-gnomad-mean
- printed value
- 37.478
- numeric value
- 37.478
- metric
- odds-ratio
- metric direction
- higher
- unit
- unitless
- metric qualifier
- rare (singletons) versus common (MAF > 5%) gnomAD variants in the tail of deleterious scores, threshold at 30 false discoveries, one-sided Fisher's exact test; all variant types, rare subsampled to the common count; mean over three random seeds; zero-shot
- source locator
- Supplementary Table 4 (PDF page 23), row 'Window size = 64', column 'gnomAD' / 'mean'
- missing metadata
- uncertainty: reason: unreported; note: Only mean and max over seeds are printed
- review
- method: pdf-text-parse; independent-cell-check; method note: A fresh 2026-10-14 download of the v2 PDF (2,247,726 bytes, SHA-256 ea7f7c3a...3389; bioRxiv stamps each download, see the source's hash_scope) was converted with pdftotext -layout. An independent parser read all 17 rows of Supplementary Table 4 and matched all 34 gnomAD cells (value, mean or max, metric, protocol). For every row, the reviewed ClinVar result with the same row label uses the same system and equals the parsed ClinVar cell, which ties this text layer and the row identities to the reviewed batch. Six cells also checked by hand.; reviewer: claude; reviewer note: Claude (Opus 5.5) research agent, independent of the extractor; no human review claimed; date: 2026-10-14; artifact sha256: ea7f7c3a1b9c2f0b4d5b03f38b68ad9f37d02e8f49b04f3184e12138c2a43389; retrieval url: https://www.biorxiv.org/content/10.1101/2023.10.10.561776v2.full.pdf; note: Extraction: Parsed the pdftotext 26.08.0 -layout text of a 2026-10-14 re-download of the v2 PDF (same length as the pinned file; SHA-256 c066dab122b359b90059a81f325fa2cd83122df14a700eae8d35021471c10978) with extract/build_gnomad.py; caption, column groups, 'mean max' header, the 17 row labels and the page asserted, and the row's ClinVar, COSMIC and OMIM cells asserted equal to the reviewed store values. Independent review 2026-10-14: matched. The 30-false-discovery threshold is from the Figure 2f caption, which the table caption says the ablation follows ('same approach'); recorded in the protocol limitations.