rewirebio.iobenchmarks
Evidence claim

GPN pretraining corpus and held-out chromosomes

Training data of GPN trained on eight Brassicales genomes, repeat loss weight 0.1 (PNAS 2023)

Eight RefSeq-annotated Brassicales reference genomes, one per genus, filtered from NCBI Genome: Arabidopsis thaliana, Brassica rapa, Camelina sativa, Capsella rubella, Carica papaya, Eutrema salsugineum, Raphanus sativus and Tarenaya hassleriana. Arabidopsis thaliana chromosomes 4 and 5 were held out for validation and testing. Windows of 512 bp at a step of 256 bp were sampled with the reverse complement, drawn from exons with a small intronic flank, promoters 1,000 bp upstream of transcription start sites and an equivalent amount of random windows, with each genome subset to the number of windows from Arabidopsis.

What the GPN paper states about the data its main model was trained on.

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.

12 evidence rows matching the loaded filters

Claims, original sources and review scope · Release 2026-11-21-c8bea25e5abb
Property and statementOriginal source and locationReview and provenance
Claimed field
training_data
Context-only references
DNA language models are powerful predictors of genome-wide variant effects (PNAS 2023)

Original source ↗

SI Appendix Table S1; Materials and Methods, 'Pretraining'

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: PNAS 120(44) e2311219120, published 26 October 2023; Europe PMC full text JATS XML of PMC10622914
Retrieved: 2026-11-16T00:29:00Z

not individually reviewed

No individual claim review recorded

Audit details

Field: attributes.field

Source artifact SHA-256: 8ff2184c07f5e6807af0b9d5103d669e1920e735ab5a4b1ea856c070077ae1e0

Hash scope: Hash scope not separately documented; inspect source record

Format: JATS XML

Claimed field
training_data
Context-only references
SI Appendix for PNAS 2023 GPN paper (Tables S1 to S3, Figures S1 to S12)

Original source ↗

SI Appendix Table S1; Materials and Methods, 'Pretraining'

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: pnas.2311219120.sapp.pdf as served for PMC10622914, 18 pages, created 2023-09-22
Retrieved: 2026-11-16T00:30:00Z

not individually reviewed

No individual claim review recorded

Audit details

Field: attributes.field

Source artifact SHA-256: e9ea2399b54c66f17b6088f205e17242521bd31fcb6b608841cdaf69cf363b16

Hash scope: pdftotext -layout (poppler 26.08.0) on the pinned PDF; the text layer is kept at extract/si-appendix.layout.txt, sha256 6ee1d0c99cc16aa2f5781755204665f0ee08a8f966a9a52c4e1e804ba3bf335a.

Format: PDF

Archive member: pnas.2311219120.sapp.pdf

Source location
SI Appendix Table S1; Materials and Methods, 'Pretraining'
Context-only references
DNA language models are powerful predictors of genome-wide variant effects (PNAS 2023)

Original source ↗

SI Appendix Table S1; Materials and Methods, 'Pretraining'

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: PNAS 120(44) e2311219120, published 26 October 2023; Europe PMC full text JATS XML of PMC10622914
Retrieved: 2026-11-16T00:29:00Z

not individually reviewed

No individual claim review recorded

Audit details

Field: attributes.source_locator

Source artifact SHA-256: 8ff2184c07f5e6807af0b9d5103d669e1920e735ab5a4b1ea856c070077ae1e0

Hash scope: Hash scope not separately documented; inspect source record

Format: JATS XML

Source location
SI Appendix Table S1; Materials and Methods, 'Pretraining'
Context-only references
SI Appendix for PNAS 2023 GPN paper (Tables S1 to S3, Figures S1 to S12)

Original source ↗

SI Appendix Table S1; Materials and Methods, 'Pretraining'

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: pnas.2311219120.sapp.pdf as served for PMC10622914, 18 pages, created 2023-09-22
Retrieved: 2026-11-16T00:30:00Z

not individually reviewed

No individual claim review recorded

Audit details

Field: attributes.source_locator

Source artifact SHA-256: e9ea2399b54c66f17b6088f205e17242521bd31fcb6b608841cdaf69cf363b16

Hash scope: pdftotext -layout (poppler 26.08.0) on the pinned PDF; the text layer is kept at extract/si-appendix.layout.txt, sha256 6ee1d0c99cc16aa2f5781755204665f0ee08a8f966a9a52c4e1e804ba3bf335a.

Format: PDF

Archive member: pnas.2311219120.sapp.pdf

Claimed value
Eight RefSeq-annotated Brassicales reference genomes, one per genus, filtered from NCBI Genome: Arabidopsis thaliana, Brassica rapa, Camelina sativa, Capsella rubella, Carica papaya, Eutrema salsugineum, Raphanus sativus and Tarenaya hassleriana. Arabidopsis thaliana chromosomes 4 and 5 were held out for validation and testing. Windows of 512 bp at a step of 256 bp were sampled with the reverse complement, drawn from exons with a small intronic flank, promoters 1,000 bp upstream of transcription start sites and an equivalent amount of random windows, with each genome subset to the number of windows from Arabidopsis.
Context-only references
DNA language models are powerful predictors of genome-wide variant effects (PNAS 2023)

Original source ↗

SI Appendix Table S1; Materials and Methods, 'Pretraining'

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: PNAS 120(44) e2311219120, published 26 October 2023; Europe PMC full text JATS XML of PMC10622914
Retrieved: 2026-11-16T00:29:00Z

not individually reviewed

No individual claim review recorded

Audit details

Field: attributes.value

Source artifact SHA-256: 8ff2184c07f5e6807af0b9d5103d669e1920e735ab5a4b1ea856c070077ae1e0

Hash scope: Hash scope not separately documented; inspect source record

Format: JATS XML

Claimed value
Eight RefSeq-annotated Brassicales reference genomes, one per genus, filtered from NCBI Genome: Arabidopsis thaliana, Brassica rapa, Camelina sativa, Capsella rubella, Carica papaya, Eutrema salsugineum, Raphanus sativus and Tarenaya hassleriana. Arabidopsis thaliana chromosomes 4 and 5 were held out for validation and testing. Windows of 512 bp at a step of 256 bp were sampled with the reverse complement, drawn from exons with a small intronic flank, promoters 1,000 bp upstream of transcription start sites and an equivalent amount of random windows, with each genome subset to the number of windows from Arabidopsis.
Context-only references
SI Appendix for PNAS 2023 GPN paper (Tables S1 to S3, Figures S1 to S12)

Original source ↗

SI Appendix Table S1; Materials and Methods, 'Pretraining'

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: pnas.2311219120.sapp.pdf as served for PMC10622914, 18 pages, created 2023-09-22
Retrieved: 2026-11-16T00:30:00Z

not individually reviewed

No individual claim review recorded

Audit details

Field: attributes.value

Source artifact SHA-256: e9ea2399b54c66f17b6088f205e17242521bd31fcb6b608841cdaf69cf363b16

Hash scope: pdftotext -layout (poppler 26.08.0) on the pinned PDF; the text layer is kept at extract/si-appendix.layout.txt, sha256 6ee1d0c99cc16aa2f5781755204665f0ee08a8f966a9a52c4e1e804ba3bf335a.

Format: PDF

Archive member: pnas.2311219120.sapp.pdf

Description
What the GPN paper states about the data its main model was trained on.
Context-only references
DNA language models are powerful predictors of genome-wide variant effects (PNAS 2023)

Original source ↗

SI Appendix Table S1; Materials and Methods, 'Pretraining'

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: PNAS 120(44) e2311219120, published 26 October 2023; Europe PMC full text JATS XML of PMC10622914
Retrieved: 2026-11-16T00:29:00Z

not individually reviewed

No individual claim review recorded

Audit details

Field: description

Source artifact SHA-256: 8ff2184c07f5e6807af0b9d5103d669e1920e735ab5a4b1ea856c070077ae1e0

Hash scope: Hash scope not separately documented; inspect source record

Format: JATS XML

Description
What the GPN paper states about the data its main model was trained on.
Context-only references
SI Appendix for PNAS 2023 GPN paper (Tables S1 to S3, Figures S1 to S12)

Original source ↗

SI Appendix Table S1; Materials and Methods, 'Pretraining'

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: pnas.2311219120.sapp.pdf as served for PMC10622914, 18 pages, created 2023-09-22
Retrieved: 2026-11-16T00:30:00Z

not individually reviewed

No individual claim review recorded

Audit details

Field: description

Source artifact SHA-256: e9ea2399b54c66f17b6088f205e17242521bd31fcb6b608841cdaf69cf363b16

Hash scope: pdftotext -layout (poppler 26.08.0) on the pinned PDF; the text layer is kept at extract/si-appendix.layout.txt, sha256 6ee1d0c99cc16aa2f5781755204665f0ee08a8f966a9a52c4e1e804ba3bf335a.

Format: PDF

Archive member: pnas.2311219120.sapp.pdf

Relationship: subject
gpn-paper-20261116-config-gpn-brassicales-repeat-weight-0-1
Context-only references
DNA language models are powerful predictors of genome-wide variant effects (PNAS 2023)

Original source ↗

SI Appendix Table S1; Materials and Methods, 'Pretraining'

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: PNAS 120(44) e2311219120, published 26 October 2023; Europe PMC full text JATS XML of PMC10622914
Retrieved: 2026-11-16T00:29:00Z

not individually reviewed

No individual claim review recorded

Audit details

Field: links:subject:gpn-paper-20261116-config-gpn-brassicales-repeat-weight-0-1

Source artifact SHA-256: 8ff2184c07f5e6807af0b9d5103d669e1920e735ab5a4b1ea856c070077ae1e0

Hash scope: Hash scope not separately documented; inspect source record

Format: JATS XML

Relationship: subject
gpn-paper-20261116-config-gpn-brassicales-repeat-weight-0-1
Context-only references
SI Appendix for PNAS 2023 GPN paper (Tables S1 to S3, Figures S1 to S12)

Original source ↗

SI Appendix Table S1; Materials and Methods, 'Pretraining'

Shared locator for this statement’s cited sources; not a separate locator for each citation.

Version: pnas.2311219120.sapp.pdf as served for PMC10622914, 18 pages, created 2023-09-22
Retrieved: 2026-11-16T00:30:00Z

not individually reviewed

No individual claim review recorded

Audit details

Field: links:subject:gpn-paper-20261116-config-gpn-brassicales-repeat-weight-0-1

Source artifact SHA-256: e9ea2399b54c66f17b6088f205e17242521bd31fcb6b608841cdaf69cf363b16

Hash scope: pdftotext -layout (poppler 26.08.0) on the pinned PDF; the text layer is kept at extract/si-appendix.layout.txt, sha256 6ee1d0c99cc16aa2f5781755204665f0ee08a8f966a9a52c4e1e804ba3bf335a.

Format: PDF

Archive member: pnas.2311219120.sapp.pdf

How to read the evidence

Sources and history

Release 2026-11-21-c8bea25e5abb · Record review: source checked

2 source records and release historyDownload this release (gzip)
Technical metadata and extraction receipts

Stable ID: gpn-paper-20261116-claim-training-data

areas
dna-genomes
contexts
research
field
training_data
review
artifact sha256: e9ea2399b54c66f17b6088f205e17242521bd31fcb6b608841cdaf69cf363b16; method: independent-cell-check; automated-source-review; source-hash-verification; method note: Independent review pass by a different agent from the extractor, with its own script (review/recheck_values.py), which does not import extract/parse_values.py and does not read extract/values.json. The published article XML and the bioRxiv details API response were re-downloaded and matched their pinned hashes; the SI Appendix PDF was re-downloaded as the member of a freshly built Europe PMC zip (a third distinct zip hash, c7a1b20e8796c4bd384ba6f0308134e32a66e40dc881b0c1e3d0201da26e8498) and the PDF member matched. Every archived gz in artifacts/ decompresses to the hash its source record pins, and the text layer in extract/ regenerates byte for byte from the archived PDF with pdftotext -layout, poppler 26.08.0. npm run batch -- check-values reproduced the extractor's figures: 6 in_cell, 17 locator_not_parsed, no overclaimed_in_row. Each claim was read against its own source_locator in the re-retrieved artifacts. Two locators did not cover the whole claim and were corrected in this pass: the comparator bound claim now also names Materials and Methods, 'Variant Effect Prediction', where phyloP x LD, abs(phyloP) x LD, phastCons x LD and the baseline LD Score are defined, and the motif claim now also names Materials and Methods, 'Motif Analysis', where the 1,000 bp promoter window and the repeat and coding exon filter are printed. The central finding was confirmed independently: the whole JATS document holds zero table-wrap and zero table elements, and the SI Appendix figure pages print no comparator value in their text layer, only axis ticks, GPN score bin edges, threshold quantiles, method legends and the per-class variant counts of the Figure S9 subplot titles.; note: The eight assemblies were checked against the SI Appendix Table S1 text layer. The accessions print with a space where the accession carries an underscore (GCF 000001735.4) because the PDF typesets them that way; the underscore in the record is the only change made to a printed string in this batch and the review accepts it as a typesetting artefact, not a transcription error.; retrieval url: https://www.ebi.ac.uk/europepmc/webservices/rest/PMC10622914/supplementaryFiles; reviewed at: 2026-11-16T01:30:00Z; reviewer: claude; reviewer note: Claude review agent, independent of the extractor; no human review claimed. Transcription was checked against the pinned artifacts; nothing was reproduced and no value was recomputed.
source locator
SI Appendix Table S1; Materials and Methods, 'Pretraining'
value
Eight RefSeq-annotated Brassicales reference genomes, one per genus, filtered from NCBI Genome: Arabidopsis thaliana, Brassica rapa, Camelina sativa, Capsella rubella, Carica papaya, Eutrema salsugineum, Raphanus sativus and Tarenaya hassleriana. Arabidopsis thaliana chromosomes 4 and 5 were held out for validation and testing. Windows of 512 bp at a step of 256 bp were sampled with the reverse complement, drawn from exons with a small intronic flank, promoters 1,000 bp upstream of transcription start sites and an equivalent amount of random windows, with each genome subset to the number of windows from Arabidopsis.
Related records

Suggest a correction