Configuration: NT-500M, frozen embeddings with a lightweight probe (Datta et al. 2026) Protocol: Nucleotide Transformer benchmark under frozen-embedding probing (Datta et al. 2026 Table 1) Dataset: Nucleotide Transformer benchmark task data (Datta et al. 2026) 0.525 MCC (NT benchmark Enhancer, frozen-embedding probe)unitless · higher
Uncertainty: SD 0.0262
Coverage: Not reported scored / Not reported eligible
Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 1 prints the same frozen value and spread (.5252+-.0262) for NT on both 'Enhancer' and 'Enhancer Types', while the fine-tuned references differ (.548 and .424). One of the two frozen cells is likely a transcription error in the source; both are kept as printed and should not support comparison until the authors clarify.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'NT', rows 'Enhancer' and 'Enhancer Types'
Table 1 and Table 4 disagree for HyenaDNA Splice Acceptor: Table 1 prints .4752±.0108 (recovery 50), while Table 4 prints 0.7308 ± 0.0033 (validation-selected) and 0.7360 ± 0.0052 (diagnostic grid-best). Five other HyenaDNA cells of Table 1 equal Table 4's grid-best values, but Splice Donor (.7247±.0087) equals the validation-selected value instead. Table 3 and Table 4 also name different grid-best readouts for HyenaDNA Splice All (MLP+Mean against MLP+Attn). Table 1 values are kept as printed.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'HyenaDNA', rows 'Splice Acceptor' and 'Splice Donor'; Table 4; Table 3, row 'Splice All'
Methods, coverage and source NT-500M on NT benchmark Enhancer (frozen probe)
gfm2-20261011-datta2026-protocol-frozen-probe
Aggregation: Not reported
Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, row 'Enhancer', column 'NT', frozen value Configuration: NT-500M, frozen embeddings with a lightweight probe (Datta et al. 2026) Protocol: Nucleotide Transformer benchmark under frozen-embedding probing (Datta et al. 2026 Table 1) Dataset: Nucleotide Transformer benchmark task data (Datta et al. 2026) 0.525 MCC (NT benchmark Enhancer Types, frozen-embedding probe)unitless · higher
Uncertainty: SD 0.0262
Coverage: Not reported scored / Not reported eligible
Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 1 prints the same frozen value and spread (.5252+-.0262) for NT on both 'Enhancer' and 'Enhancer Types', while the fine-tuned references differ (.548 and .424). One of the two frozen cells is likely a transcription error in the source; both are kept as printed and should not support comparison until the authors clarify.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'NT', rows 'Enhancer' and 'Enhancer Types'
Table 1 and Table 4 disagree for HyenaDNA Splice Acceptor: Table 1 prints .4752±.0108 (recovery 50), while Table 4 prints 0.7308 ± 0.0033 (validation-selected) and 0.7360 ± 0.0052 (diagnostic grid-best). Five other HyenaDNA cells of Table 1 equal Table 4's grid-best values, but Splice Donor (.7247±.0087) equals the validation-selected value instead. Table 3 and Table 4 also name different grid-best readouts for HyenaDNA Splice All (MLP+Mean against MLP+Attn). Table 1 values are kept as printed.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'HyenaDNA', rows 'Splice Acceptor' and 'Splice Donor'; Table 4; Table 3, row 'Splice All'
Methods, coverage and source NT-500M on NT benchmark Enhancer Types (frozen probe)
gfm2-20261011-datta2026-protocol-frozen-probe
Aggregation: Not reported
Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, row 'Enhancer Types', column 'NT', frozen value Configuration: NT-500M, frozen embeddings with a lightweight probe (Datta et al. 2026) Protocol: Nucleotide Transformer benchmark under frozen-embedding probing (Datta et al. 2026 Table 1) Dataset: Nucleotide Transformer benchmark task data (Datta et al. 2026) 0.695 MCC (NT benchmark H3, frozen-embedding probe)unitless · higher
Uncertainty: SD 0.0111
Coverage: Not reported scored / Not reported eligible
Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 1 prints the same frozen value and spread (.5252+-.0262) for NT on both 'Enhancer' and 'Enhancer Types', while the fine-tuned references differ (.548 and .424). One of the two frozen cells is likely a transcription error in the source; both are kept as printed and should not support comparison until the authors clarify.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'NT', rows 'Enhancer' and 'Enhancer Types'
Table 1 and Table 4 disagree for HyenaDNA Splice Acceptor: Table 1 prints .4752±.0108 (recovery 50), while Table 4 prints 0.7308 ± 0.0033 (validation-selected) and 0.7360 ± 0.0052 (diagnostic grid-best). Five other HyenaDNA cells of Table 1 equal Table 4's grid-best values, but Splice Donor (.7247±.0087) equals the validation-selected value instead. Table 3 and Table 4 also name different grid-best readouts for HyenaDNA Splice All (MLP+Mean against MLP+Attn). Table 1 values are kept as printed.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'HyenaDNA', rows 'Splice Acceptor' and 'Splice Donor'; Table 4; Table 3, row 'Splice All'
Methods, coverage and source NT-500M on NT benchmark H3 (frozen probe)
gfm2-20261011-datta2026-protocol-frozen-probe
Aggregation: Not reported
Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, row 'H3', column 'NT', frozen value Configuration: NT-500M, frozen embeddings with a lightweight probe (Datta et al. 2026) Protocol: Nucleotide Transformer benchmark under frozen-embedding probing (Datta et al. 2026 Table 1) Dataset: Nucleotide Transformer benchmark task data (Datta et al. 2026) 0.332 MCC (NT benchmark H3K14ac, frozen-embedding probe)unitless · higher
Uncertainty: SD 0.0068
Coverage: Not reported scored / Not reported eligible
Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 1 prints the same frozen value and spread (.5252+-.0262) for NT on both 'Enhancer' and 'Enhancer Types', while the fine-tuned references differ (.548 and .424). One of the two frozen cells is likely a transcription error in the source; both are kept as printed and should not support comparison until the authors clarify.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'NT', rows 'Enhancer' and 'Enhancer Types'
Table 1 and Table 4 disagree for HyenaDNA Splice Acceptor: Table 1 prints .4752±.0108 (recovery 50), while Table 4 prints 0.7308 ± 0.0033 (validation-selected) and 0.7360 ± 0.0052 (diagnostic grid-best). Five other HyenaDNA cells of Table 1 equal Table 4's grid-best values, but Splice Donor (.7247±.0087) equals the validation-selected value instead. Table 3 and Table 4 also name different grid-best readouts for HyenaDNA Splice All (MLP+Mean against MLP+Attn). Table 1 values are kept as printed.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'HyenaDNA', rows 'Splice Acceptor' and 'Splice Donor'; Table 4; Table 3, row 'Splice All'
Methods, coverage and source NT-500M on NT benchmark H3K14ac (frozen probe)
gfm2-20261011-datta2026-protocol-frozen-probe
Aggregation: Not reported
Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, row 'H3K14ac', column 'NT', frozen value Configuration: NT-500M, frozen embeddings with a lightweight probe (Datta et al. 2026) Protocol: Nucleotide Transformer benchmark under frozen-embedding probing (Datta et al. 2026 Table 1) Dataset: Nucleotide Transformer benchmark task data (Datta et al. 2026) 0.4 MCC (NT benchmark H3K36me3, frozen-embedding probe)unitless · higher
Uncertainty: SD 0.0160
Coverage: Not reported scored / Not reported eligible
Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 1 prints the same frozen value and spread (.5252+-.0262) for NT on both 'Enhancer' and 'Enhancer Types', while the fine-tuned references differ (.548 and .424). One of the two frozen cells is likely a transcription error in the source; both are kept as printed and should not support comparison until the authors clarify.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'NT', rows 'Enhancer' and 'Enhancer Types'
Table 1 and Table 4 disagree for HyenaDNA Splice Acceptor: Table 1 prints .4752±.0108 (recovery 50), while Table 4 prints 0.7308 ± 0.0033 (validation-selected) and 0.7360 ± 0.0052 (diagnostic grid-best). Five other HyenaDNA cells of Table 1 equal Table 4's grid-best values, but Splice Donor (.7247±.0087) equals the validation-selected value instead. Table 3 and Table 4 also name different grid-best readouts for HyenaDNA Splice All (MLP+Mean against MLP+Attn). Table 1 values are kept as printed.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'HyenaDNA', rows 'Splice Acceptor' and 'Splice Donor'; Table 4; Table 3, row 'Splice All'
Methods, coverage and source NT-500M on NT benchmark H3K36me3 (frozen probe)
gfm2-20261011-datta2026-protocol-frozen-probe
Aggregation: Not reported
Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, row 'H3K36me3', column 'NT', frozen value Configuration: NT-500M, frozen embeddings with a lightweight probe (Datta et al. 2026) Protocol: Nucleotide Transformer benchmark under frozen-embedding probing (Datta et al. 2026 Table 1) Dataset: Nucleotide Transformer benchmark task data (Datta et al. 2026) 0.328 MCC (NT benchmark H3K4me1, frozen-embedding probe)unitless · higher
Uncertainty: SD 0.0152
Coverage: Not reported scored / Not reported eligible
Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 1 prints the same frozen value and spread (.5252+-.0262) for NT on both 'Enhancer' and 'Enhancer Types', while the fine-tuned references differ (.548 and .424). One of the two frozen cells is likely a transcription error in the source; both are kept as printed and should not support comparison until the authors clarify.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'NT', rows 'Enhancer' and 'Enhancer Types'
Table 1 and Table 4 disagree for HyenaDNA Splice Acceptor: Table 1 prints .4752±.0108 (recovery 50), while Table 4 prints 0.7308 ± 0.0033 (validation-selected) and 0.7360 ± 0.0052 (diagnostic grid-best). Five other HyenaDNA cells of Table 1 equal Table 4's grid-best values, but Splice Donor (.7247±.0087) equals the validation-selected value instead. Table 3 and Table 4 also name different grid-best readouts for HyenaDNA Splice All (MLP+Mean against MLP+Attn). Table 1 values are kept as printed.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'HyenaDNA', rows 'Splice Acceptor' and 'Splice Donor'; Table 4; Table 3, row 'Splice All'
Methods, coverage and source NT-500M on NT benchmark H3K4me1 (frozen probe)
gfm2-20261011-datta2026-protocol-frozen-probe
Aggregation: Not reported
Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, row 'H3K4me1', column 'NT', frozen value Configuration: NT-500M, frozen embeddings with a lightweight probe (Datta et al. 2026) Protocol: Nucleotide Transformer benchmark under frozen-embedding probing (Datta et al. 2026 Table 1) Dataset: Nucleotide Transformer benchmark task data (Datta et al. 2026) 0.254 MCC (NT benchmark H3K4me2, frozen-embedding probe)unitless · higher
Uncertainty: SD 0.0128
Coverage: Not reported scored / Not reported eligible
Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 1 prints the same frozen value and spread (.5252+-.0262) for NT on both 'Enhancer' and 'Enhancer Types', while the fine-tuned references differ (.548 and .424). One of the two frozen cells is likely a transcription error in the source; both are kept as printed and should not support comparison until the authors clarify.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'NT', rows 'Enhancer' and 'Enhancer Types'
Table 1 and Table 4 disagree for HyenaDNA Splice Acceptor: Table 1 prints .4752±.0108 (recovery 50), while Table 4 prints 0.7308 ± 0.0033 (validation-selected) and 0.7360 ± 0.0052 (diagnostic grid-best). Five other HyenaDNA cells of Table 1 equal Table 4's grid-best values, but Splice Donor (.7247±.0087) equals the validation-selected value instead. Table 3 and Table 4 also name different grid-best readouts for HyenaDNA Splice All (MLP+Mean against MLP+Attn). Table 1 values are kept as printed.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'HyenaDNA', rows 'Splice Acceptor' and 'Splice Donor'; Table 4; Table 3, row 'Splice All'
Methods, coverage and source NT-500M on NT benchmark H3K4me2 (frozen probe)
gfm2-20261011-datta2026-protocol-frozen-probe
Aggregation: Not reported
Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, row 'H3K4me2', column 'NT', frozen value Configuration: NT-500M, frozen embeddings with a lightweight probe (Datta et al. 2026) Protocol: Nucleotide Transformer benchmark under frozen-embedding probing (Datta et al. 2026 Table 1) Dataset: Nucleotide Transformer benchmark task data (Datta et al. 2026) 0.204 MCC (NT benchmark H3K4me3, frozen-embedding probe)unitless · higher
Uncertainty: SD 0.0227
Coverage: Not reported scored / Not reported eligible
Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 1 prints the same frozen value and spread (.5252+-.0262) for NT on both 'Enhancer' and 'Enhancer Types', while the fine-tuned references differ (.548 and .424). One of the two frozen cells is likely a transcription error in the source; both are kept as printed and should not support comparison until the authors clarify.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'NT', rows 'Enhancer' and 'Enhancer Types'
Table 1 and Table 4 disagree for HyenaDNA Splice Acceptor: Table 1 prints .4752±.0108 (recovery 50), while Table 4 prints 0.7308 ± 0.0033 (validation-selected) and 0.7360 ± 0.0052 (diagnostic grid-best). Five other HyenaDNA cells of Table 1 equal Table 4's grid-best values, but Splice Donor (.7247±.0087) equals the validation-selected value instead. Table 3 and Table 4 also name different grid-best readouts for HyenaDNA Splice All (MLP+Mean against MLP+Attn). Table 1 values are kept as printed.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'HyenaDNA', rows 'Splice Acceptor' and 'Splice Donor'; Table 4; Table 3, row 'Splice All'
Methods, coverage and source NT-500M on NT benchmark H3K4me3 (frozen probe)
gfm2-20261011-datta2026-protocol-frozen-probe
Aggregation: Not reported
Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, row 'H3K4me3', column 'NT', frozen value Configuration: NT-500M, frozen embeddings with a lightweight probe (Datta et al. 2026) Protocol: Nucleotide Transformer benchmark under frozen-embedding probing (Datta et al. 2026 Table 1) Dataset: Nucleotide Transformer benchmark task data (Datta et al. 2026) 0.566 MCC (NT benchmark H3K79me3, frozen-embedding probe)unitless · higher
Uncertainty: SD 0.0085
Coverage: Not reported scored / Not reported eligible
Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 1 prints the same frozen value and spread (.5252+-.0262) for NT on both 'Enhancer' and 'Enhancer Types', while the fine-tuned references differ (.548 and .424). One of the two frozen cells is likely a transcription error in the source; both are kept as printed and should not support comparison until the authors clarify.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'NT', rows 'Enhancer' and 'Enhancer Types'
Table 1 and Table 4 disagree for HyenaDNA Splice Acceptor: Table 1 prints .4752±.0108 (recovery 50), while Table 4 prints 0.7308 ± 0.0033 (validation-selected) and 0.7360 ± 0.0052 (diagnostic grid-best). Five other HyenaDNA cells of Table 1 equal Table 4's grid-best values, but Splice Donor (.7247±.0087) equals the validation-selected value instead. Table 3 and Table 4 also name different grid-best readouts for HyenaDNA Splice All (MLP+Mean against MLP+Attn). Table 1 values are kept as printed.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'HyenaDNA', rows 'Splice Acceptor' and 'Splice Donor'; Table 4; Table 3, row 'Splice All'
Methods, coverage and source NT-500M on NT benchmark H3K79me3 (frozen probe)
gfm2-20261011-datta2026-protocol-frozen-probe
Aggregation: Not reported
Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, row 'H3K79me3', column 'NT', frozen value Configuration: NT-500M, frozen embeddings with a lightweight probe (Datta et al. 2026) Protocol: Nucleotide Transformer benchmark under frozen-embedding probing (Datta et al. 2026 Table 1) Dataset: Nucleotide Transformer benchmark task data (Datta et al. 2026) 0.399 MCC (NT benchmark H3K9ac, frozen-embedding probe)unitless · higher
Uncertainty: SD 0.0155
Coverage: Not reported scored / Not reported eligible
Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 1 prints the same frozen value and spread (.5252+-.0262) for NT on both 'Enhancer' and 'Enhancer Types', while the fine-tuned references differ (.548 and .424). One of the two frozen cells is likely a transcription error in the source; both are kept as printed and should not support comparison until the authors clarify.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'NT', rows 'Enhancer' and 'Enhancer Types'
Table 1 and Table 4 disagree for HyenaDNA Splice Acceptor: Table 1 prints .4752±.0108 (recovery 50), while Table 4 prints 0.7308 ± 0.0033 (validation-selected) and 0.7360 ± 0.0052 (diagnostic grid-best). Five other HyenaDNA cells of Table 1 equal Table 4's grid-best values, but Splice Donor (.7247±.0087) equals the validation-selected value instead. Table 3 and Table 4 also name different grid-best readouts for HyenaDNA Splice All (MLP+Mean against MLP+Attn). Table 1 values are kept as printed.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'HyenaDNA', rows 'Splice Acceptor' and 'Splice Donor'; Table 4; Table 3, row 'Splice All'
Methods, coverage and source NT-500M on NT benchmark H3K9ac (frozen probe)
gfm2-20261011-datta2026-protocol-frozen-probe
Aggregation: Not reported
Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, row 'H3K9ac', column 'NT', frozen value Configuration: NT-500M, frozen embeddings with a lightweight probe (Datta et al. 2026) Protocol: Nucleotide Transformer benchmark under frozen-embedding probing (Datta et al. 2026 Table 1) Dataset: Nucleotide Transformer benchmark task data (Datta et al. 2026) 0.714 MCC (NT benchmark H4, frozen-embedding probe)unitless · higher
Uncertainty: SD 0.0084
Coverage: Not reported scored / Not reported eligible
Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 1 prints the same frozen value and spread (.5252+-.0262) for NT on both 'Enhancer' and 'Enhancer Types', while the fine-tuned references differ (.548 and .424). One of the two frozen cells is likely a transcription error in the source; both are kept as printed and should not support comparison until the authors clarify.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'NT', rows 'Enhancer' and 'Enhancer Types'
Table 1 and Table 4 disagree for HyenaDNA Splice Acceptor: Table 1 prints .4752±.0108 (recovery 50), while Table 4 prints 0.7308 ± 0.0033 (validation-selected) and 0.7360 ± 0.0052 (diagnostic grid-best). Five other HyenaDNA cells of Table 1 equal Table 4's grid-best values, but Splice Donor (.7247±.0087) equals the validation-selected value instead. Table 3 and Table 4 also name different grid-best readouts for HyenaDNA Splice All (MLP+Mean against MLP+Attn). Table 1 values are kept as printed.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'HyenaDNA', rows 'Splice Acceptor' and 'Splice Donor'; Table 4; Table 3, row 'Splice All'
Methods, coverage and source NT-500M on NT benchmark H4 (frozen probe)
gfm2-20261011-datta2026-protocol-frozen-probe
Aggregation: Not reported
Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, row 'H4', column 'NT', frozen value Configuration: NT-500M, frozen embeddings with a lightweight probe (Datta et al. 2026) Protocol: Nucleotide Transformer benchmark under frozen-embedding probing (Datta et al. 2026 Table 1) Dataset: Nucleotide Transformer benchmark task data (Datta et al. 2026) 0.302 MCC (NT benchmark H4ac, frozen-embedding probe)unitless · higher
Uncertainty: SD 0.0000
Coverage: Not reported scored / Not reported eligible
Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 1 prints the same frozen value and spread (.5252+-.0262) for NT on both 'Enhancer' and 'Enhancer Types', while the fine-tuned references differ (.548 and .424). One of the two frozen cells is likely a transcription error in the source; both are kept as printed and should not support comparison until the authors clarify.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'NT', rows 'Enhancer' and 'Enhancer Types'
Table 1 and Table 4 disagree for HyenaDNA Splice Acceptor: Table 1 prints .4752±.0108 (recovery 50), while Table 4 prints 0.7308 ± 0.0033 (validation-selected) and 0.7360 ± 0.0052 (diagnostic grid-best). Five other HyenaDNA cells of Table 1 equal Table 4's grid-best values, but Splice Donor (.7247±.0087) equals the validation-selected value instead. Table 3 and Table 4 also name different grid-best readouts for HyenaDNA Splice All (MLP+Mean against MLP+Attn). Table 1 values are kept as printed.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'HyenaDNA', rows 'Splice Acceptor' and 'Splice Donor'; Table 4; Table 3, row 'Splice All'
Methods, coverage and source NT-500M on NT benchmark H4ac (frozen probe)
gfm2-20261011-datta2026-protocol-frozen-probe
Aggregation: Not reported
Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, row 'H4ac', column 'NT', frozen value Configuration: NT-500M, frozen embeddings with a lightweight probe (Datta et al. 2026) Protocol: Nucleotide Transformer benchmark under frozen-embedding probing (Datta et al. 2026 Table 1) Dataset: Nucleotide Transformer benchmark task data (Datta et al. 2026) 0.928 F1 (NT benchmark Promoter All, frozen-embedding probe)fraction · higher
Uncertainty: SD 0.0042
Coverage: Not reported scored / Not reported eligible
Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 1 prints the same frozen value and spread (.5252+-.0262) for NT on both 'Enhancer' and 'Enhancer Types', while the fine-tuned references differ (.548 and .424). One of the two frozen cells is likely a transcription error in the source; both are kept as printed and should not support comparison until the authors clarify.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'NT', rows 'Enhancer' and 'Enhancer Types'
Table 1 and Table 4 disagree for HyenaDNA Splice Acceptor: Table 1 prints .4752±.0108 (recovery 50), while Table 4 prints 0.7308 ± 0.0033 (validation-selected) and 0.7360 ± 0.0052 (diagnostic grid-best). Five other HyenaDNA cells of Table 1 equal Table 4's grid-best values, but Splice Donor (.7247±.0087) equals the validation-selected value instead. Table 3 and Table 4 also name different grid-best readouts for HyenaDNA Splice All (MLP+Mean against MLP+Attn). Table 1 values are kept as printed.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'HyenaDNA', rows 'Splice Acceptor' and 'Splice Donor'; Table 4; Table 3, row 'Splice All'
Methods, coverage and source NT-500M on NT benchmark Promoter All (frozen probe)
gfm2-20261011-datta2026-protocol-frozen-probe
Aggregation: Not reported
Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, row 'Promoter All', column 'NT', frozen value Configuration: NT-500M, frozen embeddings with a lightweight probe (Datta et al. 2026) Protocol: Nucleotide Transformer benchmark under frozen-embedding probing (Datta et al. 2026 Table 1) Dataset: Nucleotide Transformer benchmark task data (Datta et al. 2026) 0.937 F1 (NT benchmark Promoter Non-TATA, frozen-embedding probe)fraction · higher
Uncertainty: SD 0.0024
Coverage: Not reported scored / Not reported eligible
Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 1 prints the same frozen value and spread (.5252+-.0262) for NT on both 'Enhancer' and 'Enhancer Types', while the fine-tuned references differ (.548 and .424). One of the two frozen cells is likely a transcription error in the source; both are kept as printed and should not support comparison until the authors clarify.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'NT', rows 'Enhancer' and 'Enhancer Types'
Table 1 and Table 4 disagree for HyenaDNA Splice Acceptor: Table 1 prints .4752±.0108 (recovery 50), while Table 4 prints 0.7308 ± 0.0033 (validation-selected) and 0.7360 ± 0.0052 (diagnostic grid-best). Five other HyenaDNA cells of Table 1 equal Table 4's grid-best values, but Splice Donor (.7247±.0087) equals the validation-selected value instead. Table 3 and Table 4 also name different grid-best readouts for HyenaDNA Splice All (MLP+Mean against MLP+Attn). Table 1 values are kept as printed.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'HyenaDNA', rows 'Splice Acceptor' and 'Splice Donor'; Table 4; Table 3, row 'Splice All'
Methods, coverage and source NT-500M on NT benchmark Promoter Non-TATA (frozen probe)
gfm2-20261011-datta2026-protocol-frozen-probe
Aggregation: Not reported
Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, row 'Promoter Non-TATA', column 'NT', frozen value Configuration: NT-500M, frozen embeddings with a lightweight probe (Datta et al. 2026) Protocol: Nucleotide Transformer benchmark under frozen-embedding probing (Datta et al. 2026 Table 1) Dataset: Nucleotide Transformer benchmark task data (Datta et al. 2026) 0.918 F1 (NT benchmark Promoter TATA, frozen-embedding probe)fraction · higher
Uncertainty: SD 0.0056
Coverage: Not reported scored / Not reported eligible
Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 1 prints the same frozen value and spread (.5252+-.0262) for NT on both 'Enhancer' and 'Enhancer Types', while the fine-tuned references differ (.548 and .424). One of the two frozen cells is likely a transcription error in the source; both are kept as printed and should not support comparison until the authors clarify.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'NT', rows 'Enhancer' and 'Enhancer Types'
Table 1 and Table 4 disagree for HyenaDNA Splice Acceptor: Table 1 prints .4752±.0108 (recovery 50), while Table 4 prints 0.7308 ± 0.0033 (validation-selected) and 0.7360 ± 0.0052 (diagnostic grid-best). Five other HyenaDNA cells of Table 1 equal Table 4's grid-best values, but Splice Donor (.7247±.0087) equals the validation-selected value instead. Table 3 and Table 4 also name different grid-best readouts for HyenaDNA Splice All (MLP+Mean against MLP+Attn). Table 1 values are kept as printed.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'HyenaDNA', rows 'Splice Acceptor' and 'Splice Donor'; Table 4; Table 3, row 'Splice All'
Methods, coverage and source NT-500M on NT benchmark Promoter TATA (frozen probe)
gfm2-20261011-datta2026-protocol-frozen-probe
Aggregation: Not reported
Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, row 'Promoter TATA', column 'NT', frozen value Configuration: NT-500M, frozen embeddings with a lightweight probe (Datta et al. 2026) Protocol: Nucleotide Transformer benchmark under frozen-embedding probing (Datta et al. 2026 Table 1) Dataset: Nucleotide Transformer benchmark task data (Datta et al. 2026) 0.86 F1 (NT benchmark Splice Acceptor, frozen-embedding probe)fraction · higher
Uncertainty: SD 0.0012
Coverage: Not reported scored / Not reported eligible
Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 1 prints the same frozen value and spread (.5252+-.0262) for NT on both 'Enhancer' and 'Enhancer Types', while the fine-tuned references differ (.548 and .424). One of the two frozen cells is likely a transcription error in the source; both are kept as printed and should not support comparison until the authors clarify.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'NT', rows 'Enhancer' and 'Enhancer Types'
Table 1 and Table 4 disagree for HyenaDNA Splice Acceptor: Table 1 prints .4752±.0108 (recovery 50), while Table 4 prints 0.7308 ± 0.0033 (validation-selected) and 0.7360 ± 0.0052 (diagnostic grid-best). Five other HyenaDNA cells of Table 1 equal Table 4's grid-best values, but Splice Donor (.7247±.0087) equals the validation-selected value instead. Table 3 and Table 4 also name different grid-best readouts for HyenaDNA Splice All (MLP+Mean against MLP+Attn). Table 1 values are kept as printed.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'HyenaDNA', rows 'Splice Acceptor' and 'Splice Donor'; Table 4; Table 3, row 'Splice All'
Methods, coverage and source NT-500M on NT benchmark Splice Acceptor (frozen probe)
gfm2-20261011-datta2026-protocol-frozen-probe
Aggregation: Not reported
Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, row 'Splice Acceptor', column 'NT', frozen value Configuration: NT-500M, frozen embeddings with a lightweight probe (Datta et al. 2026) Protocol: Nucleotide Transformer benchmark under frozen-embedding probing (Datta et al. 2026 Table 1) Dataset: Nucleotide Transformer benchmark task data (Datta et al. 2026) 0.687 F1 (NT benchmark Splice All, frozen-embedding probe)fraction · higher
Uncertainty: SD 0.0083
Coverage: Not reported scored / Not reported eligible
Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 1 prints the same frozen value and spread (.5252+-.0262) for NT on both 'Enhancer' and 'Enhancer Types', while the fine-tuned references differ (.548 and .424). One of the two frozen cells is likely a transcription error in the source; both are kept as printed and should not support comparison until the authors clarify.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'NT', rows 'Enhancer' and 'Enhancer Types'
Table 1 and Table 4 disagree for HyenaDNA Splice Acceptor: Table 1 prints .4752±.0108 (recovery 50), while Table 4 prints 0.7308 ± 0.0033 (validation-selected) and 0.7360 ± 0.0052 (diagnostic grid-best). Five other HyenaDNA cells of Table 1 equal Table 4's grid-best values, but Splice Donor (.7247±.0087) equals the validation-selected value instead. Table 3 and Table 4 also name different grid-best readouts for HyenaDNA Splice All (MLP+Mean against MLP+Attn). Table 1 values are kept as printed.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'HyenaDNA', rows 'Splice Acceptor' and 'Splice Donor'; Table 4; Table 3, row 'Splice All'
Methods, coverage and source NT-500M on NT benchmark Splice All (frozen probe)
gfm2-20261011-datta2026-protocol-frozen-probe
Aggregation: Not reported
Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, row 'Splice All', column 'NT', frozen value Configuration: NT-500M, frozen embeddings with a lightweight probe (Datta et al. 2026) Protocol: Nucleotide Transformer benchmark under frozen-embedding probing (Datta et al. 2026 Table 1) Dataset: Nucleotide Transformer benchmark task data (Datta et al. 2026) 0.81 F1 (NT benchmark Splice Donor, frozen-embedding probe)fraction · higher
Uncertainty: SD 0.0054
Coverage: Not reported scored / Not reported eligible
Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 1 prints the same frozen value and spread (.5252+-.0262) for NT on both 'Enhancer' and 'Enhancer Types', while the fine-tuned references differ (.548 and .424). One of the two frozen cells is likely a transcription error in the source; both are kept as printed and should not support comparison until the authors clarify.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'NT', rows 'Enhancer' and 'Enhancer Types'
Table 1 and Table 4 disagree for HyenaDNA Splice Acceptor: Table 1 prints .4752±.0108 (recovery 50), while Table 4 prints 0.7308 ± 0.0033 (validation-selected) and 0.7360 ± 0.0052 (diagnostic grid-best). Five other HyenaDNA cells of Table 1 equal Table 4's grid-best values, but Splice Donor (.7247±.0087) equals the validation-selected value instead. Table 3 and Table 4 also name different grid-best readouts for HyenaDNA Splice All (MLP+Mean against MLP+Attn). Table 1 values are kept as printed.Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, column 'HyenaDNA', rows 'Splice Acceptor' and 'Splice Donor'; Table 4; Table 3, row 'Splice All'
Methods, coverage and source NT-500M on NT benchmark Splice Donor (frozen probe)
gfm2-20261011-datta2026-protocol-frozen-probe
Aggregation: Not reported
Frozen but Not Always Accessible: A Representation Analysis of Genomic Language Models · Table 1, row 'Splice Donor', column 'NT', frozen value