Protocol

ViroBench ALL viruses host classification, T-Split, Macro-F1

Evidence concern: excluded from comparisons

Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM'

Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)'

ViroBench host classification on ALL viruses under the temporal split by collection date; frozen embeddings with a standardised MLP head, window-level predictions aggregated to sequence level.

76 evaluations · 76 results

Overview

An explanatory profile has not yet been reviewed for this record.

Consult the linked sources for architecture or protocol details. Missing evidence is not evidence of a missing capability.

76 recorded evaluations, 76 metric rows. A comparison chart has not yet been validated for these results. The table retains the individual findings and their sources.

View coverage and remaining gaps across all benchmarks

Results

Results are available, but no reviewed comparison panel is linked in this release.

All evaluations

76 evaluations · 76 results. Different protocols are not a single leaderboard.

Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 · Dataset: ViroBench ALL viruses, T-Split test set

Sorted by Macro F1 (ViroBench ALL viruses, host, T-Split) (higher is better). The best value in each column is highlighted. Decimals are rounded for display; each value links to the printed value and its source.

Tested configuration	Macro F1 (ViroBench ALL viruses, host, T-Split)
Kraken2 (ViroBench classification)Evaluation	69.41 (0.00)% (best in column)
BLAST (ViroBench classification)Evaluation	65.55 (0.00)%
ViroDNABERT2 (ViroBench classification)Evaluation	56.73 (0.80)%
LucaVirus-default-step3.8M (ViroBench classification)Evaluation	54.84 (1.54)%
ViroCaduceus (ViroBench classification)Evaluation	50.13 (1.42)%
GenomeOcean-4B (ViroBench classification)Evaluation	48.75 (1.14)%
Evo2-7B-Base (ViroBench classification)Evaluation	48.32 (0.72)%
Evo2-7B (ViroBench classification)Evaluation	47.60 (0.29)%
LucaOne-default-step36M (ViroBench classification)Evaluation	47.52 (0.39)%
AIDO.DNA-7B (ViroBench classification)Evaluation	47.47 (1.38)%
DNABERT-S (ViroBench classification)Evaluation	47.41 (0.88)%
Evo2-40B-Base (ViroBench classification)Evaluation	47.39 (0.66)%
AIDO.DNA-300M (ViroBench classification)Evaluation	46.96 (2.85)%
Evo2-40B (ViroBench classification)Evaluation	45.71 (1.19)%
LucaVirus-gene-step3.8M (ViroBench classification)Evaluation	45.01 (2.93)%
BiLSTM (ViroBench classification)Evaluation	44.69 (1.31)%
LucaOne-gene-step36.8M (ViroBench classification)Evaluation	44.49 (1.33)%
GENA-LM-bert-Base-t2t (ViroBench classification)Evaluation	42.34 (0.41)%
AIDO.RNA-650M-CDS (ViroBench classification)Evaluation	41.85 (2.75)%
GenomeOcean-100M (ViroBench classification)Evaluation	40.27 (5.44)%
GENA-LM-bert-large-t2t (ViroBench classification)Evaluation	39.83 (4.86)%
AIDO.RNA-1.6B-CDS (ViroBench classification)Evaluation	38.19 (1.97)%
GENA-LM-bigbird-Base-t2t (ViroBench classification)Evaluation	37.92 (1.47)%
GenomeOcean-500M (ViroBench classification)Evaluation	37.67 (1.62)%
NTv3-650M-post (ViroBench classification)Evaluation	36.72 (2.26)%
MP-RNA (ViroBench classification)Evaluation	36.61 (0.11)%
Evo1.5-8K-Base (ViroBench classification)Evaluation	35.91 (1.14)%
Evo1-131K-Base (ViroBench classification)Evaluation	35.48 (2.51)%
AIDO.RNA-650M (ViroBench classification)Evaluation	33.99 (0.73)%
Evo1-8K-Base (ViroBench classification)Evaluation	33.40 (0.58)%
NTv3-100M-post (ViroBench classification)Evaluation	33.02 (1.25)%
ViroHyena-253M (ViroBench classification)Evaluation	30.70 (2.62)%
AIDO.RNA-1.6B (ViroBench classification)Evaluation	30.32 (3.20)%
NTv3-100M-pre (ViroBench classification)Evaluation	29.81 (1.32)%
RNA-FM (ViroBench classification)Evaluation	25.59 (4.37)%
GROVER (ViroBench classification)Evaluation	25.55 (0.78)%
ViroHyena-6M (ViroBench classification)Evaluation	25.31 (2.48)%
NTv2-50M-ms (ViroBench classification)Evaluation	25.20 (21.60)%
CNN (ViroBench classification)Evaluation	25.16 (5.52)%
DNABERT-6 (ViroBench classification)Evaluation	24.60 (2.21)%
NTv2-250M-ms (ViroBench classification)Evaluation	24.43 (21.13)%
NTv2-500M-ms (ViroBench classification)Evaluation	24.17 (20.84)%
RiNALMo (ViroBench classification)Evaluation	23.84 (8.19)%
NTv2-100M-ms (ViroBench classification)Evaluation	23.08 (19.87)%
Caduceus-ph (ViroBench classification)Evaluation	21.68 (4.35)%
NT-2.5b-ms (ViroBench classification)Evaluation	21.18 (3.47)%
ViroHyena-1M (ViroBench classification)Evaluation	21.04 (1.49)%
BiRNA-BERT (ViroBench classification)Evaluation	21.00 (3.82)%
ViroHyena-436K (ViroBench classification)Evaluation	19.85 (3.12)%
Caduceus-ps (ViroBench classification)Evaluation	19.54 (2.97)%
OmniReg-GPT (ViroBench classification)Evaluation	18.15 (4.24)%
HyenaDNA-tiny-16k (ViroBench classification)Evaluation	17.75 (1.82)%
DNABERT-3 (ViroBench classification)Evaluation	16.42 (4.20)%
RNABERT (ViroBench classification)Evaluation	15.98 (1.59)%
NTv2-50M-ms-3kmer (ViroBench classification)Evaluation	15.65 (13.99)%
DNABERT-5 (ViroBench classification)Evaluation	14.96 (0.87)%
Evo2-1B-Base (ViroBench classification)Evaluation	12.96 (0.30)%
HyenaDNA-small-32k (ViroBench classification)Evaluation	12.95 (11.06)%
DNABERT-4 (ViroBench classification)Evaluation	12.51 (2.83)%
HyenaDNA-medium-450k (ViroBench classification)Evaluation	11.96 (10.15)%
NT-500M-human (ViroBench classification)Evaluation	9.64 (7.81)%
DNABERT-2-117M (ViroBench classification)Evaluation	9.02 (8.07)%
NT-2.5b-1000g (ViroBench classification)Evaluation	8.21 (12.11)%
HyenaDNA-tiny-1k (ViroBench classification)Evaluation	6.44 (10.71)%
NTv3-650M-pre (ViroBench classification)Evaluation	6.21 (9.96)%
HyenaDNA-medium-160k (ViroBench classification)Evaluation	5.87 (9.73)%
HyenaDNA-large-1M (ViroBench classification)Evaluation	5.65 (8.99)%
Genos-10B (ViroBench classification)Evaluation	5.54 (9.15)%
Genos-10B-v2 (ViroBench classification)Evaluation	5.24 (8.64)%
GENERator-v2-eukaryote-3b-Base (ViroBench classification)Evaluation	0.39 (0.24)%
GENERator-v2-prokaryote-1.2b-Base (ViroBench classification)Evaluation	0.39 (0.24)%
GENERator-v2-prokaryote-3b-Base (ViroBench classification)Evaluation	0.39 (0.24)%
NT-500M-1000g (ViroBench classification)Evaluation	0.39 (0.24)%
NTv3-8M-pre (ViroBench classification)Evaluation	0.39 (0.24)%
GENERator-v2-eukaryote-1.2b-Base (ViroBench classification)Evaluation	0.25 (0.00)%
Genos-1.2B (ViroBench classification)Evaluation	0.25 (0.00)%

All 76 result rows with coverage, uncertainty and sources

Filter evaluations

Tested model or methodProtocolDatasetEvaluation setupMetricEvidence origin

Applied filters: All linked evaluations

Exact evaluated configurations and original reported results
Tested configuration	Protocol and dataset	Finding	Evidence and details
Configuration: AIDO.DNA-300M (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	46.96 (2.85)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 2.85 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source AIDO.DNA-300M on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'AIDO.DNA-300M', column 'ALL viruses / Host / T-Split' (text-layer line 2569)
Configuration: AIDO.DNA-7B (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	47.47 (1.38)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 1.38 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source AIDO.DNA-7B on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'AIDO.DNA-7B', column 'ALL viruses / Host / T-Split' (text-layer line 2570)
Configuration: AIDO.RNA-1.6B (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	30.32 (3.20)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 3.20 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source AIDO.RNA-1.6B on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'AIDO.RNA-1.6B', column 'ALL viruses / Host / T-Split' (text-layer line 2607)
Configuration: AIDO.RNA-1.6B-CDS (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	38.19 (1.97)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 1.97 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source AIDO.RNA-1.6B-CDS on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'AIDO.RNA-1.6B-CDS', column 'ALL viruses / Host / T-Split' (text-layer line 2609)
Configuration: AIDO.RNA-650M (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	33.99 (0.73)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 0.73 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source AIDO.RNA-650M on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'AIDO.RNA-650M', column 'ALL viruses / Host / T-Split' (text-layer line 2606)
Configuration: AIDO.RNA-650M-CDS (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	41.85 (2.75)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 2.75 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source AIDO.RNA-650M-CDS on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'AIDO.RNA-650M-CDS', column 'ALL viruses / Host / T-Split' (text-layer line 2608)
Configuration: BiLSTM (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	44.69 (1.31)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 1.31 Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source BiLSTM on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'BiLSTM', column 'ALL viruses / Host / T-Split' (text-layer line 2534)
Configuration: BiRNA-BERT (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	21.00 (3.82)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 3.82 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source BiRNA-BERT on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'BiRNA-BERT', column 'ALL viruses / Host / T-Split' (text-layer line 2612)
Configuration: BLAST (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	65.55 (0.00)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 0.00 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source BLAST on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'BLAST', column 'ALL viruses / Host / T-Split' (text-layer line 2532)
Configuration: Caduceus-ph (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	21.68 (4.35)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 4.35 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source Caduceus-ph on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'Caduceus-ph', column 'ALL viruses / Host / T-Split' (text-layer line 2571)
Configuration: Caduceus-ps (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	19.54 (2.97)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 2.97 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source Caduceus-ps on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'Caduceus-ps', column 'ALL viruses / Host / T-Split' (text-layer line 2572)
Configuration: CNN (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	25.16 (5.52)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 5.52 Coverage: Not reported scored / Not reported eligible	Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source CNN on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'CNN', column 'ALL viruses / Host / T-Split' (text-layer line 2535)
Configuration: DNABERT-2-117M (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	9.02 (8.07)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 8.07 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source DNABERT-2-117M on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'DNABERT-2-117M', column 'ALL viruses / Host / T-Split' (text-layer line 2582)
Configuration: DNABERT-3 (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	16.42 (4.20)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 4.20 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source DNABERT-3 on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'DNABERT-3', column 'ALL viruses / Host / T-Split' (text-layer line 2588)
Configuration: DNABERT-4 (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	12.51 (2.83)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 2.83 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source DNABERT-4 on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'DNABERT-4', column 'ALL viruses / Host / T-Split' (text-layer line 2589)
Configuration: DNABERT-5 (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	14.96 (0.87)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 0.87 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source DNABERT-5 on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'DNABERT-5', column 'ALL viruses / Host / T-Split' (text-layer line 2590)
Configuration: DNABERT-6 (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	24.60 (2.21)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 2.21 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source DNABERT-6 on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'DNABERT-6', column 'ALL viruses / Host / T-Split' (text-layer line 2591)
Configuration: DNABERT-S (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	47.41 (0.88)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 0.88 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source DNABERT-S on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'DNABERT-S', column 'ALL viruses / Host / T-Split' (text-layer line 2541)
Configuration: Evo1-131K-Base (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	35.48 (2.51)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 2.51 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source Evo1-131K-Base on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'Evo1-131K-Base', column 'ALL viruses / Host / T-Split' (text-layer line 2547)
Configuration: Evo1.5-8K-Base (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	35.91 (1.14)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 1.14 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source Evo1.5-8K-Base on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'Evo1.5-8K-Base', column 'ALL viruses / Host / T-Split' (text-layer line 2548)
Configuration: Evo1-8K-Base (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	33.40 (0.58)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 0.58 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source Evo1-8K-Base on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'Evo1-8K-Base', column 'ALL viruses / Host / T-Split' (text-layer line 2546)
Configuration: Evo2-1B-Base (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	12.96 (0.30)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 0.30 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source Evo2-1B-Base on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'Evo2-1B-Base', column 'ALL viruses / Host / T-Split' (text-layer line 2549)
Configuration: Evo2-40B (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	45.71 (1.19)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 1.19 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source Evo2-40B on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'Evo2-40B', column 'ALL viruses / Host / T-Split' (text-layer line 2553)
Configuration: Evo2-40B-Base (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	47.39 (0.66)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 0.66 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source Evo2-40B-Base on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'Evo2-40B-Base', column 'ALL viruses / Host / T-Split' (text-layer line 2552)
Configuration: Evo2-7B (ViroBench classification)	Protocol: ViroBench ALL viruses host classification, T-Split, Macro-F1 Dataset: ViroBench ALL viruses, T-Split test set	47.60 (0.29)% Macro F1 (ViroBench ALL viruses, host, T-Split) percent · higher Uncertainty: SD 0.29 Coverage: Not reported scored / Not reported eligible	Independent external evaluation · Source checked Evidence concern: excluded from comparisons Table 25 prints one Macro-F1 per model for the taxonomy columns. For 143 of the 144 model-by-split taxonomy cells on the ALL-virus set, the printed value equals the mean of the five per-rank values in Table 26 rounded to two decimals. The exception is RNA-FM under the genus-disjoint split: Table 25 prints 48.67 and the mean of the Table 26 ranks (59.96, 65.35, 37.58, 58.78, 43.82) is 53.10. One of the two tables is wrong for that row, so RNA-FM ALL taxonomy G-Split should not be used in a comparison without checking the published code.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'RNA-FM', column 'ALL viruses / Taxonomy / G-Split'; Table 26, row 'RNA-FM' Table 3 lists NT v3 under phage-specific pretraining coverage (P) and Table 9 lists it as non-viral (N). Table 25 groups the NTv3 rows under phage-specific coverage, which is what the configurations record. Pretraining coverage is the paper's main explanatory variable, so NT v3 rows should not be used to support a coverage-versus-performance claim.ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 3 row 'NT v3'; Table 9 row 'NT v3'; Table 25 section 'DNA Foundation Models (Phage-specific Coverage)' Methods, coverage and source Evo2-7B on ViroBench ALL viruses host classification (T-Split) lit-microbial-20261011-virobench-protocol-all-host-t-split Aggregation: Not reported ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Table 25, row 'Evo2-7B', column 'ALL viruses / Host / T-Split' (text-layer line 2551)

Source checking is not independent reproduction. Release 2026-10-22-1b7cf027e4f6.

Methods and evaluation design

Procedure, tasks and evaluated configurations

Recorded evaluations

Each evaluation records what was tested and under which conditions.

Explore all linked results

Baseline coverage

Reference methods help show what a model adds beyond simple controls. We track a null control and a conventional method for each protocol.

0 of 2 active baseline roles have published Rewire measurements in this release. Measurements on a selected protocol do not establish coverage of an entire suite.

No execution recipe linked to this protocol. Recipe availability does not establish a completed evaluation.

Author-reported evaluations: 8
External evaluations: 68

Literature evidence is not a Rewire measurement. Executed but unpublished runs and private review status are not included.

Null control

Proposed control: requires review

Training-set class prior where supervised fitting is permitted

Protocol-specific applicability, permitted inputs, access, split, evaluator and execution requirements need review before implementation or execution.

This is a suggested selection rule, not a validated method or a measured score.

Conventional reference

Proposed control: requires review

Regularised classifier on simple permitted features, or protocol's conventional reference

Protocol-specific applicability, permitted inputs, access, split, evaluator and execution requirements need review before implementation or execution.

This is a suggested selection rule, not a validated method or a measured score.

Protocol coverage CSV (gzip) · Model evaluation matrix (gzip) · Source table (gzip) · Release and checksums (gzip)

Coverage is derived from release 2026-10-22-1b7cf027e4f6. Source citations describe the original records; they do not validate an unreviewed baseline proposal. No results have been generated by this audit.

Run instructions

No runnable recipe has been reviewed for this protocol. Dataset access, model requirements, licences and compute requirements must be checked against its sources before execution.

Strengths, limitations and unresolved questions

Evidence

Source checking verifies the cited claim or transcription. It does not establish independent reproduction.

Evidence table

Inspect claims, sources and review details

Trace each statement to its source and review. A context-only reference supports the record generally; it does not verify an individual field. Source checking does not reproduce an experiment.

One row per statement and cited source. Multiple citations are not independent evaluations. Shared locators are labelled explicitly.

0 evidence rows matching the loaded filters

Claims, original sources and review scope · Release 2026-10-22-1b7cf027e4f6
Property and statement	Original source and location	Review and provenance

No evidence rows match these filters. Choose another scope or clear the search.

How to read the evidence

Sources and history

Release 2026-10-22-1b7cf027e4f6 · Record review: source checked

1 source record and release history

ViroBench: Benchmarking Nucleotide Foundation Models on Viral Genomics Tasks · Original source · arXiv 2605.25388v1 (submitted 2026-05-25); 42-page PDF; KDD '26 header on every page

Download this release (gzip)

Technical metadata and extraction receipts

Stable ID: lit-microbial-20261011-virobench-protocol-all-host-t-split

areas: microbes-communities; dna-genomes
contexts: research
protocol: ViroBench classification, ALL viruses, Host, T-Split; score is Macro-F1 as printed in Table 25
version: arXiv 2605.25388v1
metric: macro-f1
metric direction: higher
unit: percent
unit detail: Macro-F1 x100 as printed
limitations: The benchmark authors ran every model, including their own in-house ViroHyena, ViroDNABERT2 and ViroCaduceus models and the BiLSTM and CNN baselines.; Results are the mean (standard deviation) across hyperparameter settings (window sizes 512, 1024 and 2048 with 8, 4 and 2 sampled windows, and learning rates 1e-2 to 1e-4), not across random seeds (Section 4.1.2).; Values are printed without a unit; the ablation tables state Macro-F1 in %, so they are recorded as percent.; Host classification assigns each virus to exactly one coarse-grained host category; the source says many viruses have multi-host tropism and that single-label framing can penalise biologically reasonable secondary host associations (Appendix limitations discussion, text-layer lines 2185 to 2199).
source locator: Table 25, column group 'ALL viruses', 'Host', 'T-Split'

Related records

part of: ViroBench
uses data: ViroBench ALL viruses, T-Split test set
assessment: AIDO.DNA-300M on ViroBench ALL viruses host classification (T-Split)
assessment: AIDO.DNA-7B on ViroBench ALL viruses host classification (T-Split)
assessment: AIDO.RNA-1.6B on ViroBench ALL viruses host classification (T-Split)
assessment: AIDO.RNA-1.6B-CDS on ViroBench ALL viruses host classification (T-Split)
assessment: AIDO.RNA-650M on ViroBench ALL viruses host classification (T-Split)
assessment: AIDO.RNA-650M-CDS on ViroBench ALL viruses host classification (T-Split)
assessment: BiLSTM on ViroBench ALL viruses host classification (T-Split)
assessment: BiRNA-BERT on ViroBench ALL viruses host classification (T-Split)
assessment: BLAST on ViroBench ALL viruses host classification (T-Split)
assessment: Caduceus-ph on ViroBench ALL viruses host classification (T-Split)
assessment: Caduceus-ps on ViroBench ALL viruses host classification (T-Split)
assessment: CNN on ViroBench ALL viruses host classification (T-Split)
assessment: DNABERT-2-117M on ViroBench ALL viruses host classification (T-Split)
assessment: DNABERT-3 on ViroBench ALL viruses host classification (T-Split)
assessment: DNABERT-4 on ViroBench ALL viruses host classification (T-Split)
assessment: DNABERT-5 on ViroBench ALL viruses host classification (T-Split)
assessment: DNABERT-6 on ViroBench ALL viruses host classification (T-Split)
assessment: DNABERT-S on ViroBench ALL viruses host classification (T-Split)
assessment: Evo1-131K-Base on ViroBench ALL viruses host classification (T-Split)
assessment: Evo1.5-8K-Base on ViroBench ALL viruses host classification (T-Split)
assessment: Evo1-8K-Base on ViroBench ALL viruses host classification (T-Split)
assessment: Evo2-1B-Base on ViroBench ALL viruses host classification (T-Split)
assessment: Evo2-40B on ViroBench ALL viruses host classification (T-Split)
assessment: Evo2-40B-Base on ViroBench ALL viruses host classification (T-Split)
assessment: Evo2-7B on ViroBench ALL viruses host classification (T-Split)
assessment: Evo2-7B-Base on ViroBench ALL viruses host classification (T-Split)
assessment: GENA-LM-bert-Base-t2t on ViroBench ALL viruses host classification (T-Split)
assessment: GENA-LM-bert-large-t2t on ViroBench ALL viruses host classification (T-Split)
assessment: GENA-LM-bigbird-Base-t2t on ViroBench ALL viruses host classification (T-Split)
assessment: GENERator-v2-eukaryote-1.2b-Base on ViroBench ALL viruses host classification (T-Split)
assessment: GENERator-v2-eukaryote-3b-Base on ViroBench ALL viruses host classification (T-Split)
assessment: GENERator-v2-prokaryote-1.2b-Base on ViroBench ALL viruses host classification (T-Split)
assessment: GENERator-v2-prokaryote-3b-Base on ViroBench ALL viruses host classification (T-Split)
assessment: GenomeOcean-100M on ViroBench ALL viruses host classification (T-Split)
assessment: GenomeOcean-4B on ViroBench ALL viruses host classification (T-Split)
assessment: GenomeOcean-500M on ViroBench ALL viruses host classification (T-Split)
assessment: Genos-1.2B on ViroBench ALL viruses host classification (T-Split)
assessment: Genos-10B on ViroBench ALL viruses host classification (T-Split)
assessment: Genos-10B-v2 on ViroBench ALL viruses host classification (T-Split)
assessment: GROVER on ViroBench ALL viruses host classification (T-Split)
assessment: HyenaDNA-large-1M on ViroBench ALL viruses host classification (T-Split)
assessment: HyenaDNA-medium-160k on ViroBench ALL viruses host classification (T-Split)
assessment: HyenaDNA-medium-450k on ViroBench ALL viruses host classification (T-Split)
assessment: HyenaDNA-small-32k on ViroBench ALL viruses host classification (T-Split)
assessment: HyenaDNA-tiny-16k on ViroBench ALL viruses host classification (T-Split)
assessment: HyenaDNA-tiny-1k on ViroBench ALL viruses host classification (T-Split)
assessment: Kraken2 on ViroBench ALL viruses host classification (T-Split)
assessment: LucaOne-default-step36M on ViroBench ALL viruses host classification (T-Split)
assessment: LucaOne-gene-step36.8M on ViroBench ALL viruses host classification (T-Split)
assessment: LucaVirus-default-step3.8M on ViroBench ALL viruses host classification (T-Split)
assessment: LucaVirus-gene-step3.8M on ViroBench ALL viruses host classification (T-Split)
assessment: MP-RNA on ViroBench ALL viruses host classification (T-Split)
assessment: NT-2.5b-1000g on ViroBench ALL viruses host classification (T-Split)
assessment: NT-2.5b-ms on ViroBench ALL viruses host classification (T-Split)
assessment: NT-500M-1000g on ViroBench ALL viruses host classification (T-Split)
assessment: NT-500M-human on ViroBench ALL viruses host classification (T-Split)
assessment: NTv2-100M-ms on ViroBench ALL viruses host classification (T-Split)
assessment: NTv2-250M-ms on ViroBench ALL viruses host classification (T-Split)
assessment: NTv2-500M-ms on ViroBench ALL viruses host classification (T-Split)
assessment: NTv2-50M-ms-3kmer on ViroBench ALL viruses host classification (T-Split)
assessment: NTv2-50M-ms on ViroBench ALL viruses host classification (T-Split)
assessment: NTv3-100M-post on ViroBench ALL viruses host classification (T-Split)
assessment: NTv3-100M-pre on ViroBench ALL viruses host classification (T-Split)
assessment: NTv3-650M-post on ViroBench ALL viruses host classification (T-Split)
assessment: NTv3-650M-pre on ViroBench ALL viruses host classification (T-Split)
assessment: NTv3-8M-pre on ViroBench ALL viruses host classification (T-Split)
assessment: OmniReg-GPT on ViroBench ALL viruses host classification (T-Split)
assessment: RiNALMo on ViroBench ALL viruses host classification (T-Split)
assessment: RNA-FM on ViroBench ALL viruses host classification (T-Split)
assessment: RNABERT on ViroBench ALL viruses host classification (T-Split)
assessment: ViroCaduceus on ViroBench ALL viruses host classification (T-Split)
assessment: ViroDNABERT2 on ViroBench ALL viruses host classification (T-Split)
assessment: ViroHyena-1M on ViroBench ALL viruses host classification (T-Split)
assessment: ViroHyena-253M on ViroBench ALL viruses host classification (T-Split)
assessment: ViroHyena-436K on ViroBench ALL viruses host classification (T-Split)
assessment: ViroHyena-6M on ViroBench ALL viruses host classification (T-Split)

Suggest a correction