| Configuration: Bacformer (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUPRC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 50.09 ± 22.00 AUPRC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 22.00 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceBacformer on BacBench essential gene prediction, linear probe AUPRC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'Bacformer', column 'AUPRC' |
|---|
| Configuration: Bacformer (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUROC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 79.49 ± 10.48 AUROC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 10.48 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceBacformer on BacBench essential gene prediction, linear probe AUROC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'Bacformer', column 'AUROC' |
|---|
| Configuration: Bacformer Large (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUPRC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 55.87 ± 21.38 AUPRC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 21.38 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceBacformer Large on BacBench essential gene prediction, linear probe AUPRC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'Bacformer Large', column 'AUPRC' |
|---|
| Configuration: Bacformer Large (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUROC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 77.97 ± 11.16 AUROC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 11.16 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceBacformer Large on BacBench essential gene prediction, linear probe AUROC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'Bacformer Large', column 'AUROC' |
|---|
| Configuration: DNABERT-2 (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUPRC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 30.21 ± 19.81 AUPRC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 19.81 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceDNABERT-2 on BacBench essential gene prediction, linear probe AUPRC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'DNABERT-2', column 'AUPRC' |
|---|
| Configuration: DNABERT-2 (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUROC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 64.97 ± 8.29 AUROC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 8.29 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceDNABERT-2 on BacBench essential gene prediction, linear probe AUROC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'DNABERT-2', column 'AUROC' |
|---|
| Configuration: ESM-2 (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUPRC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 43.98 ± 23.17 AUPRC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 23.17 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESM-2 on BacBench essential gene prediction, linear probe AUPRC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'ESM-2', column 'AUPRC' |
|---|
| Configuration: ESM-2 (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUROC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 80.40 ± 10.15 AUROC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 10.15 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESM-2 on BacBench essential gene prediction, linear probe AUROC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'ESM-2', column 'AUROC' |
|---|
| Configuration: ESM-C (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUPRC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 58.20 ± 21.86 AUPRC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 21.86 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESM-C on BacBench essential gene prediction, linear probe AUPRC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'ESM-C', column 'AUPRC' |
|---|
| Configuration: ESM-C (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUROC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 81.18 ± 10.88 AUROC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 10.88 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESM-C on BacBench essential gene prediction, linear probe AUROC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'ESM-C', column 'AUROC' |
|---|
| Configuration: Evo-2 (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUPRC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 15.37 ± 23.71 AUPRC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 23.71 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo-2 on BacBench essential gene prediction, linear probe AUPRC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'Evo-2', column 'AUPRC' |
|---|
| Configuration: Evo-2 (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUROC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 56.05 ± 9.46 AUROC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 9.46 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo-2 on BacBench essential gene prediction, linear probe AUROC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'Evo-2', column 'AUROC' |
|---|
| Configuration: Evo (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUPRC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 13.67 ± 22.40 AUPRC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 22.40 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo on BacBench essential gene prediction, linear probe AUPRC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'Evo', column 'AUPRC' |
|---|
| Configuration: Evo (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUROC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 51.78 ± 4.16 AUROC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 4.16 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo on BacBench essential gene prediction, linear probe AUROC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'Evo', column 'AUROC' |
|---|
| Configuration: gLM2 (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUPRC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 64.14 ± 20.69 AUPRC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 20.69 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourcegLM2 on BacBench essential gene prediction, linear probe AUPRC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'gLM2', column 'AUPRC' |
|---|
| Configuration: gLM2 (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUROC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 80.57 ± 12.19 AUROC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 12.19 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourcegLM2 on BacBench essential gene prediction, linear probe AUROC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'gLM2', column 'AUROC' |
|---|
| Configuration: Mistral-DNA (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUPRC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 14.93 ± 23.20 AUPRC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 23.20 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceMistral-DNA on BacBench essential gene prediction, linear probe AUPRC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'Mistral-DNA', column 'AUPRC' |
|---|
| Configuration: Mistral-DNA (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUROC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 58.84 ± 3.82 AUROC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 3.82 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceMistral-DNA on BacBench essential gene prediction, linear probe AUROC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'Mistral-DNA', column 'AUROC' |
|---|
| Configuration: Nucleotide Transformer (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUPRC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 25.96 ± 21.26 AUPRC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 21.26 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceNucleotide Transformer on BacBench essential gene prediction, linear probe AUPRC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'Nucleotide Transformer', column 'AUPRC' |
|---|
| Configuration: Nucleotide Transformer (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUROC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 65.68 ± 8.02 AUROC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 8.02 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceNucleotide Transformer on BacBench essential gene prediction, linear probe AUROC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'Nucleotide Transformer', column 'AUROC' |
|---|
| Configuration: ProkBERT (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUPRC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 41.06 ± 20.91 AUPRC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 20.91 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceProkBERT on BacBench essential gene prediction, linear probe AUPRC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'ProkBERT', column 'AUPRC' |
|---|
| Configuration: ProkBERT (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUROC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 74.57 ± 7.25 AUROC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 7.25 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceProkBERT on BacBench essential gene prediction, linear probe AUROC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'ProkBERT', column 'AUROC' |
|---|
| Configuration: ProtBERT (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUPRC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 50.82 ± 22.42 AUPRC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 22.42 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceProtBERT on BacBench essential gene prediction, linear probe AUPRC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'ProtBERT', column 'AUPRC' |
|---|
| Configuration: ProtBERT (frozen embeddings with a trained linear head) | Protocol: BacBench essential gene prediction, linear probe AUROC Dataset: BacBench essential gene dataset from the Database of Essential Genes | 82.45 ± 10.58 AUROC (linear head on frozen gene embeddings, genus-level split, median over held-out test genomes) unitless · higher Uncertainty: SD 10.58 Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceProtBERT on BacBench essential gene prediction, linear probe AUROC The backbone is frozen and a single linear layer, preceded by dropout of 0.2 and layer normalisation, is trained on the gene embeddings with binary cross-entropy loss, at most 100 epochs, early stopping patience 10 on macro AUROC over validation genomes, Adam with weight decay 0.02, and a learning-rate sweep per model. Evo and Evo-2 are reported without layer normalisation, which the authors state worked better for them. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S14, row 'ProtBERT', column 'AUROC' |
|---|