| Configuration: Bacformer (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUPRC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 50.31 ± 26.61 AUPRC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 26.61 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceBacformer on BacBench operon identification, zero-shot AUPRC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'Bacformer', column 'AUPRC' |
|---|
| Configuration: Bacformer (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUROC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 77.59 ± 6.64 AUROC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 6.64 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceBacformer on BacBench operon identification, zero-shot AUROC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'Bacformer', column 'AUROC' |
|---|
| Configuration: Bacformer Large (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUPRC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 49.98 ± 26.92 AUPRC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 26.92 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceBacformer Large on BacBench operon identification, zero-shot AUPRC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'Bacformer Large', column 'AUPRC' |
|---|
| Configuration: Bacformer Large (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUROC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 77.00 ± 7.46 AUROC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 7.46 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceBacformer Large on BacBench operon identification, zero-shot AUROC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'Bacformer Large', column 'AUROC' |
|---|
| Configuration: DNABERT-2 (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUPRC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 47.53 ± 30.07 AUPRC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 30.07 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceDNABERT-2 on BacBench operon identification, zero-shot AUPRC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'DNABERT-2', column 'AUPRC' |
|---|
| Configuration: DNABERT-2 (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUROC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 74.13 ± 10.14 AUROC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 10.14 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceDNABERT-2 on BacBench operon identification, zero-shot AUROC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'DNABERT-2', column 'AUROC' |
|---|
| Configuration: ESM-2 (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUPRC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 46.27 ± 29.55 AUPRC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 29.55 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESM-2 on BacBench operon identification, zero-shot AUPRC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'ESM-2', column 'AUPRC' |
|---|
| Configuration: ESM-2 (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUROC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 73.02 ± 10.30 AUROC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 10.30 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESM-2 on BacBench operon identification, zero-shot AUROC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'ESM-2', column 'AUROC' |
|---|
| Configuration: ESM-C (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUPRC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 47.79 ± 29.39 AUPRC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 29.39 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESM-C on BacBench operon identification, zero-shot AUPRC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'ESM-C', column 'AUPRC' |
|---|
| Configuration: ESM-C (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUROC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 75.25 ± 9.26 AUROC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 9.26 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESM-C on BacBench operon identification, zero-shot AUROC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'ESM-C', column 'AUROC' |
|---|
| Configuration: Evo-2 (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUPRC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 45.01 ± 29.14 AUPRC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 29.14 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo-2 on BacBench operon identification, zero-shot AUPRC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'Evo-2', column 'AUPRC' |
|---|
| Configuration: Evo-2 (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUROC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 71.63 ± 10.31 AUROC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 10.31 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo-2 on BacBench operon identification, zero-shot AUROC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'Evo-2', column 'AUROC' |
|---|
| Configuration: Evo (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUPRC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 48.21 ± 28.27 AUPRC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 28.27 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo on BacBench operon identification, zero-shot AUPRC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'Evo', column 'AUPRC' |
|---|
| Configuration: Evo (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUROC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 74.77 ± 8.33 AUROC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 8.33 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo on BacBench operon identification, zero-shot AUROC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'Evo', column 'AUROC' |
|---|
| Configuration: gLM2 (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUPRC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 45.86 ± 30.18 AUPRC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 30.18 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourcegLM2 on BacBench operon identification, zero-shot AUPRC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'gLM2', column 'AUPRC' |
|---|
| Configuration: gLM2 (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUROC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 72.85 ± 10.77 AUROC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 10.77 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourcegLM2 on BacBench operon identification, zero-shot AUROC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'gLM2', column 'AUROC' |
|---|
| Configuration: Mistral-DNA (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUPRC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 47.48 ± 29.47 AUPRC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 29.47 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceMistral-DNA on BacBench operon identification, zero-shot AUPRC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'Mistral-DNA', column 'AUPRC' |
|---|
| Configuration: Mistral-DNA (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUROC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 74.16 ± 9.70 AUROC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 9.70 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceMistral-DNA on BacBench operon identification, zero-shot AUROC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'Mistral-DNA', column 'AUROC' |
|---|
| Configuration: Nucleotide Transformer (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUPRC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 48.19 ± 27.47 AUPRC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 27.47 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceNucleotide Transformer on BacBench operon identification, zero-shot AUPRC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'Nucleotide Transformer', column 'AUPRC' |
|---|
| Configuration: Nucleotide Transformer (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUROC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 74.50 ± 8.57 AUROC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 8.57 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceNucleotide Transformer on BacBench operon identification, zero-shot AUROC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'Nucleotide Transformer', column 'AUROC' |
|---|
| Configuration: ProkBERT (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUPRC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 49.03 ± 30.07 AUPRC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 30.07 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceProkBERT on BacBench operon identification, zero-shot AUPRC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'ProkBERT', column 'AUPRC' |
|---|
| Configuration: ProkBERT (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUROC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 75.77 ± 9.69 AUROC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 9.69 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceProkBERT on BacBench operon identification, zero-shot AUROC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'ProkBERT', column 'AUROC' |
|---|
| Configuration: ProtBERT (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUPRC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 47.58 ± 27.98 AUPRC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 27.98 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceProtBERT on BacBench operon identification, zero-shot AUPRC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'ProtBERT', column 'AUPRC' |
|---|
| Configuration: ProtBERT (frozen embeddings, no trained head) | Protocol: BacBench operon identification, zero-shot AUROC Dataset: BacBench operon identification dataset from long-read RNA sequencing | 75.11 ± 8.65 AUROC (zero-shot operon boundary prediction scored by embedding cosine similarity times a same-strand indicator, mean over the five sequenced strains) unitless · higher Uncertainty: SD 8.65 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceProtBERT on BacBench operon identification, zero-shot AUROC Zero-shot. For each pair of contiguous genes the score is the cosine similarity between the two gene embeddings multiplied by an indicator that both genes are on the same strand; the intergenic-distance term of the Bacformer operon model is deliberately left out so that the comparison is about the embeddings. Performance is averaged across all five sequenced strains. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S1, row 'ProtBERT', column 'AUROC' |
|---|