BioBench Atlas
Nucleic-acid results
Benchmarks/B03

Tier B · benchmark_suite

Numeric results available

BEND

Human-genome tasks with biologically aligned splits; inspect each track for leakage controls.

B03community_protocolnot_applicable

Result coverage

105 result rows across 7 protocols

Primary sources
primary_sources_screened
Related works
priority_systematic_search_complete
Authority readiness
verified_for_internal_use
Literature cutoff
2026-08-06
TrackNumeric coverageEvidence closureBlocker
B03-TR01
BEND — TF binding-site prediction		numeric_results_available
0 results		closed_no_eligible_public_numeric_resultrelated_work_or_configuration_fields_incomplete
B03-TR02
BEND — Chromatin accessibility prediction		numeric_results_available
15 results		closed_with_numeric_evidenceNR
B03-TR03
BEND — Splice-site prediction		numeric_results_available
0 results		closed_no_eligible_public_numeric_resultrelated_work_or_configuration_fields_incomplete
B03-TR04
BEND — Sequence-to-gene-expression prediction		numeric_results_available
0 results		closed_no_eligible_public_numeric_resultrelated_work_or_configuration_fields_incomplete
B03-TR05
BEND — Regulatory variant-effect prediction		numeric_results_available
0 results		closed_no_eligible_public_numeric_resultrelated_work_or_configuration_fields_incomplete
B03-TR06
BEND — Evolutionary constraint and conservation prediction		numeric_results_available
0 results		closed_no_eligible_public_numeric_resultrelated_work_or_configuration_fields_incomplete

Protocol-level results

Choose the evaluation unit before comparing scores.

Each card keeps the four claim types inside one protocol + metric fingerprint.

Genomic element classification · NR · same_source_table

gene_finding

Full results →

gene finding · official 4780/597/597 split · frozen embeddings with two-layer CNN probe

Matthews correlation coefficientclassification · 16 results

Official baseline

AUGUSTUS;ResNet;CNN0.8 / 0.46 / 0Matthews correlation coefficient · fractionEvidence ↗

Original-table best

AUGUSTUS0.8Matthews correlation coefficient · fractionEvidence ↗

Official board / challenge leader

No eligible claimNo qualifying result exists in this snapshot.

Strict cross-work SOTA

No eligible claimAn original-table best is never substituted for cross-work SOTA.

Fingerprint: PMF-v1-2001C5DAC5342AD09874405E

Enhancer recognition · NR · same_source_table

enhancer_annotation

Full results →

enhancer annotation · official 10-fold cross-validation · frozen embeddings with two-layer CNN probe

AUPRCclassification · 16 results

Official baseline

Enformer;ResNet;CNN0.07 / 0.06 / 0.03AUPRC · fractionEvidence ↗

Original-table best

Enformer0.07AUPRC · fractionEvidence ↗

Official board / challenge leader

No eligible claimNo qualifying result exists in this snapshot.

Strict cross-work SOTA

No eligible claimAn original-table best is never substituted for cross-work SOTA.

Fingerprint: PMF-v1-63582BC0CE29C29D670CCA0E

Chromatin accessibility prediction · B03-TR02 · same_source_table

chromatin_accessibility

Full results →

chromatin accessibility · official 1,354,042/279,422/372,153 split · frozen embeddings with two-layer CNN probe

AUROCclassification · 15 results

Official baseline

Basset;CNN0.85 / 0.75AUROC · fractionEvidence ↗

Original-table best

DNABERT;Basset0.85 / 0.85AUROC · fractionEvidence ↗

Official board / challenge leader

No eligible claimNo qualifying result exists in this snapshot.

Strict cross-work SOTA

No eligible claimAn original-table best is never substituted for cross-work SOTA.

Fingerprint: PMF-v1-B15E30D210C84253E53BE6A6

Histone-mark prediction · NR · same_source_table

histone_modification

Full results →

histone modification · official 420,713/70,801/120,567 split · frozen embeddings with two-layer CNN probe

AUROCclassification · 15 results

Official baseline

Basset;CNN0.74 / 0.76AUROC · fractionEvidence ↗

Official board / challenge leader

No eligible claimNo qualifying result exists in this snapshot.

Strict cross-work SOTA

No eligible claimAn original-table best is never substituted for cross-work SOTA.

Fingerprint: PMF-v1-BA35048F9C6AF65802BAAA72

Histone-mark prediction · NR · same_source_table

cpg_methylation

Full results →

cpg methylation · official 743,095/109,717/106,227 split · frozen embeddings with two-layer CNN probe

AUROCclassification · 15 results

Official baseline

Basset;CNN0.93 / 0.84AUROC · fractionEvidence ↗

Official board / challenge leader

No eligible claimNo qualifying result exists in this snapshot.

Strict cross-work SOTA

No eligible claimAn original-table best is never substituted for cross-work SOTA.

Fingerprint: PMF-v1-1E401DE43ECBC985317C68B6

Expression variant-effect prediction · NR · same_source_table

variant_effects_expression

Full results →

variant effects expression · official zero-shot expression-variant set · zero-shot embedding-distance scoring

AUROCclassification · 14 results

Official board / challenge leader

No eligible claimNo qualifying result exists in this snapshot.

Strict cross-work SOTA

No eligible claimAn original-table best is never substituted for cross-work SOTA.

Fingerprint: PMF-v1-B838B136D378A81132FCA2AB

Molecular functional or pathogenic variant interpretation · NR · same_source_table

variant_effects_disease

Full results →

variant effects disease · official zero-shot disease-variant set · zero-shot embedding-distance scoring

AUROCclassification · 14 results

Official board / challenge leader

No eligible claimNo qualifying result exists in this snapshot.

Strict cross-work SOTA

No eligible claimAn original-table best is never substituted for cross-work SOTA.

Fingerprint: PMF-v1-AD84C40ADFF10C6BC431D8F9