| Configuration: Bacformer Large (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression Pearson correlation Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 65.73 ± 12.88 Pearson (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 12.88. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceBacformer Large on BacBench antibiotic resistance, MIC regression Pearson correlation One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'Bacformer Large', column 'Regression Pearson' |
|---|
| Configuration: Bacformer Large (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression R squared Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 42.59 ± 19.54 R² (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 19.54. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceBacformer Large on BacBench antibiotic resistance, MIC regression R squared One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'Bacformer Large', column 'Regression R2' |
|---|
| Configuration: Bacformer (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression Pearson correlation Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 57.53 ± 15.43 Pearson (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 15.43. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceBacformer on BacBench antibiotic resistance, MIC regression Pearson correlation One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'Bacformer', column 'Regression Pearson' |
|---|
| Configuration: Bacformer (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression R squared Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 33.84 ± 18.60 R² (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 18.60. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceBacformer on BacBench antibiotic resistance, MIC regression R squared One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'Bacformer', column 'Regression R2' |
|---|
| Configuration: DNABERT-2 (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression Pearson correlation Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 45.89 ± 19.27 Pearson (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 19.27. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceDNABERT-2 on BacBench antibiotic resistance, MIC regression Pearson correlation One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'DNABERT-2', column 'Regression Pearson' |
|---|
| Configuration: DNABERT-2 (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression R squared Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 23.61 ± 18.33 R² (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 18.33. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceDNABERT-2 on BacBench antibiotic resistance, MIC regression R squared One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'DNABERT-2', column 'Regression R2' |
|---|
| Configuration: ESM-2 (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression Pearson correlation Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 54.60 ± 16.15 Pearson (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 16.15. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESM-2 on BacBench antibiotic resistance, MIC regression Pearson correlation One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'ESM-2', column 'Regression Pearson' |
|---|
| Configuration: ESM-2 (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression R squared Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 31.18 ± 17.39 R² (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 17.39. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESM-2 on BacBench antibiotic resistance, MIC regression R squared One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'ESM-2', column 'Regression R2' |
|---|
| Configuration: ESM-C (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression Pearson correlation Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 56.79 ± 15.46 Pearson (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 15.46. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESM-C on BacBench antibiotic resistance, MIC regression Pearson correlation One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'ESM-C', column 'Regression Pearson' |
|---|
| Configuration: ESM-C (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression R squared Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 33.15 ± 17.62 R² (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 17.62. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceESM-C on BacBench antibiotic resistance, MIC regression R squared One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'ESM-C', column 'Regression R2' |
|---|
| Configuration: Evo-2 (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression Pearson correlation Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 00.53 ± 00.72 Pearson (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 00.72. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo-2 on BacBench antibiotic resistance, MIC regression Pearson correlation One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'Evo-2', column 'Regression Pearson' |
|---|
| Configuration: Evo-2 (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression R squared Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 03.95 ± 21.04 R² (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 21.04. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo-2 on BacBench antibiotic resistance, MIC regression R squared One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'Evo-2', column 'Regression R2' |
|---|
| Configuration: Evo (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression Pearson correlation Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 29.18 ± 17.26 Pearson (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 17.26. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo on BacBench antibiotic resistance, MIC regression Pearson correlation One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'Evo', column 'Regression Pearson' |
|---|
| Configuration: Evo (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression R squared Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 10.01 ± 10.06 R² (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 10.06. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceEvo on BacBench antibiotic resistance, MIC regression R squared One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'Evo', column 'Regression R2' |
|---|
| Configuration: gLM2 (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression Pearson correlation Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 49.41 ± 17.14 Pearson (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 17.14. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourcegLM2 on BacBench antibiotic resistance, MIC regression Pearson correlation One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'gLM2', column 'Regression Pearson' |
|---|
| Configuration: gLM2 (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression R squared Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 26.15 ± 18.13 R² (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 18.13. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourcegLM2 on BacBench antibiotic resistance, MIC regression R squared One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'gLM2', column 'Regression R2' |
|---|
| Configuration: Mistral-DNA (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression Pearson correlation Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 40.95 ± 21.51 Pearson (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 21.51. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceMistral-DNA on BacBench antibiotic resistance, MIC regression Pearson correlation One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'Mistral-DNA', column 'Regression Pearson' |
|---|
| Configuration: Mistral-DNA (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression R squared Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 19.71 ± 17.37 R² (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 17.37. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceMistral-DNA on BacBench antibiotic resistance, MIC regression R squared One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'Mistral-DNA', column 'Regression R2' |
|---|
| Configuration: Nucleotide Transformer (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression Pearson correlation Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 51.19 ± 16.76 Pearson (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 16.76. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceNucleotide Transformer on BacBench antibiotic resistance, MIC regression Pearson correlation One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'Nucleotide Transformer', column 'Regression Pearson' |
|---|
| Configuration: Nucleotide Transformer (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression R squared Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 27.74 ± 17.80 R² (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 17.80. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceNucleotide Transformer on BacBench antibiotic resistance, MIC regression R squared One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'Nucleotide Transformer', column 'Regression R2' |
|---|
| Configuration: ProkBERT (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression Pearson correlation Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 51.24 ± 16.91 Pearson (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 16.91. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceProkBERT on BacBench antibiotic resistance, MIC regression Pearson correlation One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'ProkBERT', column 'Regression Pearson' |
|---|
| Configuration: ProkBERT (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression R squared Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 28.00 ± 17.57 R² (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 17.57. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceProkBERT on BacBench antibiotic resistance, MIC regression R squared One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'ProkBERT', column 'Regression R2' |
|---|
| Configuration: ProtBERT (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression Pearson correlation Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 51.56 ± 17.61 Pearson (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 17.61. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceProtBERT on BacBench antibiotic resistance, MIC regression Pearson correlation One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'ProtBERT', column 'Regression Pearson' |
|---|
| Configuration: ProtBERT (frozen embeddings with a trained linear head) | Protocol: BacBench antibiotic resistance, MIC regression R squared Dataset subset: Antibiotic resistance MIC regression subset, 56 antibiotics | 28.23 ± 18.46 R² (linear head per drug on frozen genome embeddings predicting log(1+MIC), random 70/10/20 split, mean over 56 antibiotics, printed as a percentage) unitless · higher Uncertainty: SD 18.46. The caption says the spread is across antibiotics; the Methods state 37 antibiotics in the binary setting and 56 in the regression setting. Coverage: Not reported scored / Not reported eligible | Author-reported evaluation · Source checkedMethods, coverage and sourceProtBERT on BacBench antibiotic resistance, MIC regression R squared One linear model per drug on mean-pooled whole-genome embeddings: a single layer preceded by dropout of 0.1 and layer normalisation, binary cross-entropy for the resistant or susceptible setting and mean squared error for the log(1+MIC) regression, at most 100 epochs, early stopping patience 10 on validation AUROC or validation R squared, a learning-rate sweep per model across seeds and drugs, Adam with weight decay 0.01, and three random splits. Aggregation: Not reported Bacformer preprint, bioRxiv 2025.07.20.665723 version 3 · Table S18, row 'ProtBERT', column 'Regression R2' |
|---|