Configuration: Caduceus-PS 131k, 256 model dimensions, 16 layers Protocol: ClinVar pathogenic versus benign classification from likelihood ratios, AUROC by variant class Dataset subset: ClinVar Downstream of Transcript variants, at least one review star 0.72 AUROC (ClinVar pathogenic or likely pathogenic versus benign or likely benign, zero-shot allele log likelihood ratio, class Downstream of Transcript)unitless · higher
Uncertainty: Not reported by the source: Table 1 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 ClinVar Downstream of Transcript variants, likelihood-ratio scoring
phylogpn-20261116-protocol-clinvar-llr-auroc
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 1, row 'Downstream of Transcript', column 'Caduceus' Configuration: Caduceus-PS 131k, 256 model dimensions, 16 layers Protocol: ClinVar pathogenic versus benign classification from likelihood ratios, AUROC by variant class Dataset subset: ClinVar Intronic variants, at least one review star 0.64 AUROC (ClinVar pathogenic or likely pathogenic versus benign or likely benign, zero-shot allele log likelihood ratio, class Intronic)unitless · higher
Uncertainty: Not reported by the source: Table 1 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 ClinVar Intronic variants, likelihood-ratio scoring
phylogpn-20261116-protocol-clinvar-llr-auroc
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 1, row 'Intronic', column 'Caduceus' Configuration: Caduceus-PS 131k, 256 model dimensions, 16 layers Protocol: ClinVar pathogenic versus benign classification from likelihood ratios, AUROC by variant class Dataset subset: ClinVar Missense variants, at least one review star 0.55 AUROC (ClinVar pathogenic or likely pathogenic versus benign or likely benign, zero-shot allele log likelihood ratio, class Missense)unitless · higher
Uncertainty: Not reported by the source: Table 1 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 ClinVar Missense variants, likelihood-ratio scoring
phylogpn-20261116-protocol-clinvar-llr-auroc
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 1, row 'Missense', column 'Caduceus' Configuration: Caduceus-PS 131k, 256 model dimensions, 16 layers Protocol: ClinVar pathogenic versus benign classification from likelihood ratios, AUROC by variant class Dataset subset: ClinVar Noncoding variants, at least one review star 0.59 AUROC (ClinVar pathogenic or likely pathogenic versus benign or likely benign, zero-shot allele log likelihood ratio, class Noncoding)unitless · higher
Uncertainty: Not reported by the source: Table 1 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 ClinVar Noncoding variants, likelihood-ratio scoring
phylogpn-20261116-protocol-clinvar-llr-auroc
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 1, row 'Noncoding', column 'Caduceus' Configuration: Caduceus-PS 131k, 256 model dimensions, 16 layers Protocol: ClinVar pathogenic versus benign classification from likelihood ratios, AUROC by variant class Dataset subset: ClinVar Splice Acceptor variants, at least one review star 0.61 AUROC (ClinVar pathogenic or likely pathogenic versus benign or likely benign, zero-shot allele log likelihood ratio, class Splice Acceptor)unitless · higher
Uncertainty: Not reported by the source: Table 1 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 ClinVar Splice Acceptor variants, likelihood-ratio scoring
phylogpn-20261116-protocol-clinvar-llr-auroc
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 1, row 'Splice Acceptor', column 'Caduceus' Configuration: Caduceus-PS 131k, 256 model dimensions, 16 layers Protocol: ClinVar pathogenic versus benign classification from likelihood ratios, AUROC by variant class Dataset subset: ClinVar Splice Donor variants, at least one review star 0.68 AUROC (ClinVar pathogenic or likely pathogenic versus benign or likely benign, zero-shot allele log likelihood ratio, class Splice Donor)unitless · higher
Uncertainty: Not reported by the source: Table 1 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 ClinVar Splice Donor variants, likelihood-ratio scoring
phylogpn-20261116-protocol-clinvar-llr-auroc
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 1, row 'Splice Donor', column 'Caduceus' Configuration: Caduceus-PS 131k, 256 model dimensions, 16 layers Protocol: ClinVar pathogenic versus benign classification from likelihood ratios, AUROC by variant class Dataset subset: ClinVar Start Codon variants, at least one review star 0.41 AUROC (ClinVar pathogenic or likely pathogenic versus benign or likely benign, zero-shot allele log likelihood ratio, class Start Codon)unitless · higher
Uncertainty: Not reported by the source: Table 1 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 ClinVar Start Codon variants, likelihood-ratio scoring
phylogpn-20261116-protocol-clinvar-llr-auroc
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 1, row 'Start Codon', column 'Caduceus' Configuration: Caduceus-PS 131k, 256 model dimensions, 16 layers Protocol: ClinVar pathogenic versus benign classification from likelihood ratios, AUROC by variant class Dataset subset: ClinVar Stop Lost variants, at least one review star 0.43 AUROC (ClinVar pathogenic or likely pathogenic versus benign or likely benign, zero-shot allele log likelihood ratio, class Stop Lost)unitless · higher
Uncertainty: Not reported by the source: Table 1 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 ClinVar Stop Lost variants, likelihood-ratio scoring
phylogpn-20261116-protocol-clinvar-llr-auroc
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 1, row 'Stop Lost', column 'Caduceus' Configuration: Caduceus-PS 131k, 256 model dimensions, 16 layers Protocol: ClinVar pathogenic versus benign classification from likelihood ratios, AUROC by variant class Dataset subset: ClinVar Synonymous variants, at least one review star 0.55 AUROC (ClinVar pathogenic or likely pathogenic versus benign or likely benign, zero-shot allele log likelihood ratio, class Synonymous)unitless · higher
Uncertainty: Not reported by the source: Table 1 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 ClinVar Synonymous variants, likelihood-ratio scoring
phylogpn-20261116-protocol-clinvar-llr-auroc
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 1, row 'Synonymous', column 'Caduceus' Configuration: Caduceus-PS 131k, 256 model dimensions, 16 layers Protocol: ClinVar pathogenic versus benign classification from likelihood ratios, AUROC by variant class Dataset subset: ClinVar Upstream of Transcript variants, at least one review star 0.4 AUROC (ClinVar pathogenic or likely pathogenic versus benign or likely benign, zero-shot allele log likelihood ratio, class Upstream of Transcript)unitless · higher
Uncertainty: Not reported by the source: Table 1 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 ClinVar Upstream of Transcript variants, likelihood-ratio scoring
phylogpn-20261116-protocol-clinvar-llr-auroc
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 1, row 'Upstream of Transcript', column 'Caduceus' Configuration: Caduceus-PS 131k, 256 model dimensions, 16 layers Protocol: ClinVar pathogenic versus benign classification from likelihood ratios, AUROC by variant class Dataset subset: ClinVar 3’ UTR variants, at least one review star 0.63 AUROC (ClinVar pathogenic or likely pathogenic versus benign or likely benign, zero-shot allele log likelihood ratio, class 3’ UTR)unitless · higher
Uncertainty: Not reported by the source: Table 1 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 ClinVar 3’ UTR variants, likelihood-ratio scoring
phylogpn-20261116-protocol-clinvar-llr-auroc
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 1, row '3’ UTR', column 'Caduceus' Configuration: Caduceus-PS 131k, 256 model dimensions, 16 layers Protocol: ClinVar pathogenic versus benign classification from likelihood ratios, AUROC by variant class Dataset subset: ClinVar 5’ UTR variants, at least one review star 0.61 AUROC (ClinVar pathogenic or likely pathogenic versus benign or likely benign, zero-shot allele log likelihood ratio, class 5’ UTR)unitless · higher
Uncertainty: Not reported by the source: Table 1 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 ClinVar 5’ UTR variants, likelihood-ratio scoring
phylogpn-20261116-protocol-clinvar-llr-auroc
Aggregation: Not reported
A Phylogenetic Approach to Genomic Language Modeling · Table 1, row '5’ UTR', column 'Caduceus'