Method: AWD-LSTM Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.53 AUROCunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source AWD-LSTM on BEND VARIANT-EXPRESSION: Noncoding variant effects on expression
A downstream head trained on frozen embeddings, except for the expert methods and the fully supervised baselines, which are trained end to end. Metric and splits are from Table 1.
Aggregation: Not reported
BEND: Benchmarking DNA Language Models on Biologically Meaningful Tasks · Table 3, row(AWD-LSTM), column(Noncoding variant effects on expression) Configuration: DEEPSEA Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.7 AUROCunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source DEEPSEA on BEND VARIANT-EXPRESSION: Noncoding variant effects on expression
A downstream head trained on frozen embeddings, except for the expert methods and the fully supervised baselines, which are trained end to end. Metric and splits are from Table 1.
Aggregation: Not reported
BEND: Benchmarking DNA Language Models on Biologically Meaningful Tasks · Table 3, row(Expert method), column(Noncoding variant effects on expression) Configuration: DNABERT-2 Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.49 AUROCunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source DNABERT-2 on BEND VARIANT-EXPRESSION: Noncoding variant effects on expression
A downstream head trained on frozen embeddings, except for the expert methods and the fully supervised baselines, which are trained end to end. Metric and splits are from Table 1.
Aggregation: Not reported
BEND: Benchmarking DNA Language Models on Biologically Meaningful Tasks · Table 3, row(DNABERT-2), column(Noncoding variant effects on expression) Configuration: DNABERT Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.6 AUROCunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source DNABERT on BEND VARIANT-EXPRESSION: Noncoding variant effects on expression
A downstream head trained on frozen embeddings, except for the expert methods and the fully supervised baselines, which are trained end to end. Metric and splits are from Table 1.
Aggregation: Not reported
BEND: Benchmarking DNA Language Models on Biologically Meaningful Tasks · Table 3, row(DNABERT), column(Noncoding variant effects on expression) Configuration: GENA-LM BERT Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.49 AUROCunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source GENA-LM BERT on BEND VARIANT-EXPRESSION: Noncoding variant effects on expression
A downstream head trained on frozen embeddings, except for the expert methods and the fully supervised baselines, which are trained end to end. Metric and splits are from Table 1.
Aggregation: Not reported
BEND: Benchmarking DNA Language Models on Biologically Meaningful Tasks · Table 3, row(GENA-LM BERT), column(Noncoding variant effects on expression) Configuration: GENA-LM BigBird Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.49 AUROCunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source GENA-LM BigBird on BEND VARIANT-EXPRESSION: Noncoding variant effects on expression
A downstream head trained on frozen embeddings, except for the expert methods and the fully supervised baselines, which are trained end to end. Metric and splits are from Table 1.
Aggregation: Not reported
BEND: Benchmarking DNA Language Models on Biologically Meaningful Tasks · Table 3, row(GENA-LM BigBird), column(Noncoding variant effects on expression) Configuration: GROVER Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.56 AUROCunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source GROVER on BEND VARIANT-EXPRESSION: Noncoding variant effects on expression
A downstream head trained on frozen embeddings, except for the expert methods and the fully supervised baselines, which are trained end to end. Metric and splits are from Table 1.
Aggregation: Not reported
BEND: Benchmarking DNA Language Models on Biologically Meaningful Tasks · Table 3, row(GROVER), column(Noncoding variant effects on expression) Configuration: HyenaDNA large Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.51 AUROCunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source HyenaDNA large on BEND VARIANT-EXPRESSION: Noncoding variant effects on expression
A downstream head trained on frozen embeddings, except for the expert methods and the fully supervised baselines, which are trained end to end. Metric and splits are from Table 1.
Aggregation: Not reported
BEND: Benchmarking DNA Language Models on Biologically Meaningful Tasks · Table 3, row(HyenaDNA large), column(Noncoding variant effects on expression) Configuration: HyenaDNA tiny Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.47 AUROCunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source HyenaDNA tiny on BEND VARIANT-EXPRESSION: Noncoding variant effects on expression
A downstream head trained on frozen embeddings, except for the expert methods and the fully supervised baselines, which are trained end to end. Metric and splits are from Table 1.
Aggregation: Not reported
BEND: Benchmarking DNA Language Models on Biologically Meaningful Tasks · Table 3, row(HyenaDNA tiny), column(Noncoding variant effects on expression) Configuration: NT-1000G Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.45 AUROCunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source NT-1000G on BEND VARIANT-EXPRESSION: Noncoding variant effects on expression
A downstream head trained on frozen embeddings, except for the expert methods and the fully supervised baselines, which are trained end to end. Metric and splits are from Table 1.
Aggregation: Not reported
BEND: Benchmarking DNA Language Models on Biologically Meaningful Tasks · Table 3, row(NT-1000G), column(Noncoding variant effects on expression) Method: NT-H Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.55 AUROCunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source NT-H on BEND VARIANT-EXPRESSION: Noncoding variant effects on expression
A downstream head trained on frozen embeddings, except for the expert methods and the fully supervised baselines, which are trained end to end. Metric and splits are from Table 1.
Aggregation: Not reported
BEND: Benchmarking DNA Language Models on Biologically Meaningful Tasks · Table 3, row(NT-H), column(Noncoding variant effects on expression) Method: NT-MS Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.54 AUROCunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source NT-MS on BEND VARIANT-EXPRESSION: Noncoding variant effects on expression
A downstream head trained on frozen embeddings, except for the expert methods and the fully supervised baselines, which are trained end to end. Metric and splits are from Table 1.
Aggregation: Not reported
BEND: Benchmarking DNA Language Models on Biologically Meaningful Tasks · Table 3, row(NT-MS), column(Noncoding variant effects on expression) Configuration: NT-V2 Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.48 AUROCunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source NT-V2 on BEND VARIANT-EXPRESSION: Noncoding variant effects on expression
A downstream head trained on frozen embeddings, except for the expert methods and the fully supervised baselines, which are trained end to end. Metric and splits are from Table 1.
Aggregation: Not reported
BEND: Benchmarking DNA Language Models on Biologically Meaningful Tasks · Table 3, row(NT-V2), column(Noncoding variant effects on expression) Method: ResNet-LM Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.55 AUROCunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source ResNet-LM on BEND VARIANT-EXPRESSION: Noncoding variant effects on expression
A downstream head trained on frozen embeddings, except for the expert methods and the fully supervised baselines, which are trained end to end. Metric and splits are from Table 1.
Aggregation: Not reported
BEND: Benchmarking DNA Language Models on Biologically Meaningful Tasks · Table 3, row(ResNet-LM), column(Noncoding variant effects on expression) Configuration: Caduceus, largest reverse-complement-equivariant model, frozen embeddings Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.53 AUROC (BEND Expression VEP, cosine distance between the embeddings of the reference and alternative sequences)unitless · higher
Uncertainty: Not reported by the source: Table 2 prints no interval or spread.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Table 2's GENA-LM Chromatin Accessibility cell prints 0.84. The caption's rule is 'We report the best result among models in each family' and Section 4.2 says the row is reproduced from Marin et al. (BEND), but BEND's main results table prints 0.76 for GENA-LM BERT and 0.82 for GENA-LM BigBird in both the ICLR 2024 proceedings PDF pinned as evidence-expansion-bend-final-f709b6be and the arXiv rendition arXiv:2311.12570 retrieved on 2026-11-16. No subset of the GENA-LM family reproduces all seven printed cells of the row, while the DNABERT, HyenaDNA and Nucleotide Transformer rows each have one that does, so the printed value is not the family best this paper says it reports and its origin is unknown. The printed value is kept as published; this entry withholds that one result from automatic comparison. The row's other six cells and the rest of the source are not covered.A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'GENA-LM', column 'Chromatin Accessibility (AUROC)'
Methods, coverage and source Caduceus on BEND Expression VEP
bend-task-variant-expression
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'Caduceus', column 'Expression VEP (AUROC)' Configuration: DNABERT, best model of the family on each BEND task Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.6 AUROC (BEND Expression VEP, cosine distance between the embeddings of the reference and alternative sequences; best result among the models of this family)unitless · higher
Uncertainty: Not reported by the source: Table 2 prints no interval or spread.
Coverage: Not reported scored / Not reported eligible
Result quoted from another source · Source checked Evidence concern: excluded from comparisons Table 2's GENA-LM Chromatin Accessibility cell prints 0.84. The caption's rule is 'We report the best result among models in each family' and Section 4.2 says the row is reproduced from Marin et al. (BEND), but BEND's main results table prints 0.76 for GENA-LM BERT and 0.82 for GENA-LM BigBird in both the ICLR 2024 proceedings PDF pinned as evidence-expansion-bend-final-f709b6be and the arXiv rendition arXiv:2311.12570 retrieved on 2026-11-16. No subset of the GENA-LM family reproduces all seven printed cells of the row, while the DNABERT, HyenaDNA and Nucleotide Transformer rows each have one that does, so the printed value is not the family best this paper says it reports and its origin is unknown. The printed value is kept as published; this entry withholds that one result from automatic comparison. The row's other six cells and the rest of the source are not covered.A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'GENA-LM', column 'Chromatin Accessibility (AUROC)'
Methods, coverage and source DNABERT on BEND Expression VEP
bend-task-variant-expression
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'DNABERT', column 'Expression VEP (AUROC)' Configuration: DEEPSEA Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.7 AUROC (BEND Expression VEP, task-specific expert model trained end to end)unitless · higher
Uncertainty: Not reported by the source: Table 2 prints no interval or spread.
Coverage: Not reported scored / Not reported eligible
Result quoted from another source · Source checked Evidence concern: excluded from comparisons Table 2's GENA-LM Chromatin Accessibility cell prints 0.84. The caption's rule is 'We report the best result among models in each family' and Section 4.2 says the row is reproduced from Marin et al. (BEND), but BEND's main results table prints 0.76 for GENA-LM BERT and 0.82 for GENA-LM BigBird in both the ICLR 2024 proceedings PDF pinned as evidence-expansion-bend-final-f709b6be and the arXiv rendition arXiv:2311.12570 retrieved on 2026-11-16. No subset of the GENA-LM family reproduces all seven printed cells of the row, while the DNABERT, HyenaDNA and Nucleotide Transformer rows each have one that does, so the printed value is not the family best this paper says it reports and its origin is unknown. The printed value is kept as published; this entry withholds that one result from automatic comparison. The row's other six cells and the rest of the source are not covered.A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'GENA-LM', column 'Chromatin Accessibility (AUROC)'
Methods, coverage and source DeepSEA on BEND Expression VEP
bend-task-variant-expression
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'Expert Method', column 'Expression VEP (AUROC)' Configuration: GENA-LM, best model of the family on each BEND task Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.49 AUROC (BEND Expression VEP, cosine distance between the embeddings of the reference and alternative sequences; best result among the models of this family)unitless · higher
Uncertainty: Not reported by the source: Table 2 prints no interval or spread.
Coverage: Not reported scored / Not reported eligible
Result quoted from another source · Source checked Evidence concern: excluded from comparisons Table 2's GENA-LM Chromatin Accessibility cell prints 0.84. The caption's rule is 'We report the best result among models in each family' and Section 4.2 says the row is reproduced from Marin et al. (BEND), but BEND's main results table prints 0.76 for GENA-LM BERT and 0.82 for GENA-LM BigBird in both the ICLR 2024 proceedings PDF pinned as evidence-expansion-bend-final-f709b6be and the arXiv rendition arXiv:2311.12570 retrieved on 2026-11-16. No subset of the GENA-LM family reproduces all seven printed cells of the row, while the DNABERT, HyenaDNA and Nucleotide Transformer rows each have one that does, so the printed value is not the family best this paper says it reports and its origin is unknown. The printed value is kept as published; this entry withholds that one result from automatic comparison. The row's other six cells and the rest of the source are not covered.A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'GENA-LM', column 'Chromatin Accessibility (AUROC)'
Methods, coverage and source GENA-LM on BEND Expression VEP
bend-task-variant-expression
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'GENA-LM', column 'Expression VEP (AUROC)' Configuration: GPN-MSA, concatenated 128 bp window embeddings Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.44 AUROC (BEND Expression VEP, cosine distance between the embeddings of the reference and alternative sequences)unitless · higher
Uncertainty: Not reported by the source: Table 2 prints no interval or spread.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Table 2's GENA-LM Chromatin Accessibility cell prints 0.84. The caption's rule is 'We report the best result among models in each family' and Section 4.2 says the row is reproduced from Marin et al. (BEND), but BEND's main results table prints 0.76 for GENA-LM BERT and 0.82 for GENA-LM BigBird in both the ICLR 2024 proceedings PDF pinned as evidence-expansion-bend-final-f709b6be and the arXiv rendition arXiv:2311.12570 retrieved on 2026-11-16. No subset of the GENA-LM family reproduces all seven printed cells of the row, while the DNABERT, HyenaDNA and Nucleotide Transformer rows each have one that does, so the printed value is not the family best this paper says it reports and its origin is unknown. The printed value is kept as published; this entry withholds that one result from automatic comparison. The row's other six cells and the rest of the source are not covered.A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'GENA-LM', column 'Chromatin Accessibility (AUROC)'
Methods, coverage and source GPN-MSA on BEND Expression VEP
bend-task-variant-expression
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'GPN-MSA', column 'Expression VEP (AUROC)' Configuration: GPN-MSA, allele likelihood ratios on variant-effect tasks Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.56 AUROC (BEND Expression VEP, log likelihood ratio between the reference and alternative alleles)unitless · higher
Uncertainty: Not reported by the source: Table 2 prints no interval or spread.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Table 2's GENA-LM Chromatin Accessibility cell prints 0.84. The caption's rule is 'We report the best result among models in each family' and Section 4.2 says the row is reproduced from Marin et al. (BEND), but BEND's main results table prints 0.76 for GENA-LM BERT and 0.82 for GENA-LM BigBird in both the ICLR 2024 proceedings PDF pinned as evidence-expansion-bend-final-f709b6be and the arXiv rendition arXiv:2311.12570 retrieved on 2026-11-16. No subset of the GENA-LM family reproduces all seven printed cells of the row, while the DNABERT, HyenaDNA and Nucleotide Transformer rows each have one that does, so the printed value is not the family best this paper says it reports and its origin is unknown. The printed value is kept as published; this entry withholds that one result from automatic comparison. The row's other six cells and the rest of the source are not covered.A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'GENA-LM', column 'Chromatin Accessibility (AUROC)'
Methods, coverage and source GPN-MSA (LLR) on BEND Expression VEP
bend-task-variant-expression
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'GPN-MSA (LLR)', column 'Expression VEP (AUROC)' Configuration: HyenaDNA, best model of the family on each BEND task Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.51 AUROC (BEND Expression VEP, cosine distance between the embeddings of the reference and alternative sequences; best result among the models of this family)unitless · higher
Uncertainty: Not reported by the source: Table 2 prints no interval or spread.
Coverage: Not reported scored / Not reported eligible
Result quoted from another source · Source checked Evidence concern: excluded from comparisons Table 2's GENA-LM Chromatin Accessibility cell prints 0.84. The caption's rule is 'We report the best result among models in each family' and Section 4.2 says the row is reproduced from Marin et al. (BEND), but BEND's main results table prints 0.76 for GENA-LM BERT and 0.82 for GENA-LM BigBird in both the ICLR 2024 proceedings PDF pinned as evidence-expansion-bend-final-f709b6be and the arXiv rendition arXiv:2311.12570 retrieved on 2026-11-16. No subset of the GENA-LM family reproduces all seven printed cells of the row, while the DNABERT, HyenaDNA and Nucleotide Transformer rows each have one that does, so the printed value is not the family best this paper says it reports and its origin is unknown. The printed value is kept as published; this entry withholds that one result from automatic comparison. The row's other six cells and the rest of the source are not covered.A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'GENA-LM', column 'Chromatin Accessibility (AUROC)'
Methods, coverage and source HyenaDNA on BEND Expression VEP
bend-task-variant-expression
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'HyenaDNA', column 'Expression VEP (AUROC)' Configuration: Nucleotide Transformer, best model of the family on each BEND task Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.54 AUROC (BEND Expression VEP, cosine distance between the embeddings of the reference and alternative sequences; best result among the models of this family)unitless · higher
Uncertainty: Not reported by the source: Table 2 prints no interval or spread.
Coverage: Not reported scored / Not reported eligible
Result quoted from another source · Source checked Evidence concern: excluded from comparisons Table 2's GENA-LM Chromatin Accessibility cell prints 0.84. The caption's rule is 'We report the best result among models in each family' and Section 4.2 says the row is reproduced from Marin et al. (BEND), but BEND's main results table prints 0.76 for GENA-LM BERT and 0.82 for GENA-LM BigBird in both the ICLR 2024 proceedings PDF pinned as evidence-expansion-bend-final-f709b6be and the arXiv rendition arXiv:2311.12570 retrieved on 2026-11-16. No subset of the GENA-LM family reproduces all seven printed cells of the row, while the DNABERT, HyenaDNA and Nucleotide Transformer rows each have one that does, so the printed value is not the family best this paper says it reports and its origin is unknown. The printed value is kept as published; this entry withholds that one result from automatic comparison. The row's other six cells and the rest of the source are not covered.A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'GENA-LM', column 'Chromatin Accessibility (AUROC)'
Methods, coverage and source Nucleotide Transformer on BEND Expression VEP
bend-task-variant-expression
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'Nucleotide Transformer', column 'Expression VEP (AUROC)' Configuration: PhyloGPN (481 bp receptive field, about 83M parameters) Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.50* AUROC (BEND Expression VEP, cosine distance between the embeddings of the reference and alternative sequences)unitless · higher
Uncertainty: Not reported by the source: Table 2 prints no interval or spread.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Table 2's GENA-LM Chromatin Accessibility cell prints 0.84. The caption's rule is 'We report the best result among models in each family' and Section 4.2 says the row is reproduced from Marin et al. (BEND), but BEND's main results table prints 0.76 for GENA-LM BERT and 0.82 for GENA-LM BigBird in both the ICLR 2024 proceedings PDF pinned as evidence-expansion-bend-final-f709b6be and the arXiv rendition arXiv:2311.12570 retrieved on 2026-11-16. No subset of the GENA-LM family reproduces all seven printed cells of the row, while the DNABERT, HyenaDNA and Nucleotide Transformer rows each have one that does, so the printed value is not the family best this paper says it reports and its origin is unknown. The printed value is kept as published; this entry withholds that one result from automatic comparison. The row's other six cells and the rest of the source are not covered.A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'GENA-LM', column 'Chromatin Accessibility (AUROC)'
Methods, coverage and source PhyloGPN on BEND Expression VEP
bend-task-variant-expression
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'PhyloGPN', column 'Expression VEP (AUROC)' Configuration: PhyloGPN-X (pooled embeddings, 6,001 bp effective receptive field) Task: BEND VARIANT-EXPRESSION: Noncoding variant effects on expression Dataset subset: DeepSEA expression variants (BEND split) 0.53 AUROC (BEND Expression VEP, cosine distance between the embeddings of the reference and alternative sequences)unitless · higher
Uncertainty: Not reported by the source: Table 2 prints no interval or spread.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Table 2's GENA-LM Chromatin Accessibility cell prints 0.84. The caption's rule is 'We report the best result among models in each family' and Section 4.2 says the row is reproduced from Marin et al. (BEND), but BEND's main results table prints 0.76 for GENA-LM BERT and 0.82 for GENA-LM BigBird in both the ICLR 2024 proceedings PDF pinned as evidence-expansion-bend-final-f709b6be and the arXiv rendition arXiv:2311.12570 retrieved on 2026-11-16. No subset of the GENA-LM family reproduces all seven printed cells of the row, while the DNABERT, HyenaDNA and Nucleotide Transformer rows each have one that does, so the printed value is not the family best this paper says it reports and its origin is unknown. The printed value is kept as published; this entry withholds that one result from automatic comparison. The row's other six cells and the rest of the source are not covered.A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'GENA-LM', column 'Chromatin Accessibility (AUROC)'
Methods, coverage and source PhyloGPN-X on BEND Expression VEP
bend-task-variant-expression
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 2, row 'PhyloGPN-X', column 'Expression VEP (AUROC)'