- Rating
- 3(+1)
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
CoBALT jointly learns antigen–antibody binding-site prediction (BSP) and binding-affinity ranking (BAR). It constructs contact and energy-like fields from partner-conditioned residue representations combining sequence and geometric information. Alternating training uses separate BSP and BAR datasets without requiring both annotations for the same pair.
Critical strengths
-
The framework addresses a meaningful data limitation: binding-site and affinity annotations are rarely available for the same pair. The shared contact–energy representation connects residue-level interface prediction with candidate-level affinity ranking.
-
The reported improvements are promising, particularly for epitope prediction. In Table 1, epitope AUPRC increases from 56.22% to 65.60%, BAR Spearman from 0.65 to 0.70, and P@1 from 50.77% to 56.99%.
Critical weaknesses
-
BSP performance with predicted structures remains unverified. BSP uses structures from experimentally determined complexes, whereas BAR uses independently predicted monomer structures from ESMFold and IgFold (Appendix A.3). Performance on new pairs whose structures must be predicted therefore remains unclear. Please re-evaluate the existing BSP test sets using the same structure-prediction pipeline as BAR, keeping model checkpoints and experimental interface labels fixed. Report changes in epitope and paratope AUPRC, AUROC, and F1 relative to experimental-structure inputs. Or you can re-train this task using predicted structures for both training and testing.
-
The ablations do not establish the stability of component contributions. Table 2 reports only fold 1 of the random split. The single-task and component controls are useful, but their gains may depend on the split; the BAR Spearman improvement over the single-task model is only 0.68 to 0.70. Please report the key ablations across all five folds, including per-fold results and mean ± standard deviation, and repeat them under the existing 75%/80% antigen-homology-aware splits with comparable training budgets and model-selection procedures.
Questions for authors
-
Why are only antigen-homology-aware splits reported? Could you evaluate unseen antibody families and a stricter split that holds out both antigen and antibody clusters? Please specify the antibody similarity criterion and coordinate these partitions across BSP and BAR.
-
Why is AbLWR cited but omitted from the baselines? It also uses attention among candidates and ListMLE. Please explain the omission and, where feasible, provide a comparison with matched data, splits, and candidate-list settings. If implementation or protocol constraints prevent this, please clarify them.
8 comments
Response for Question1
Q1: BSP performance with predicted structures remains unverified.
We thank the reviewer for raising this important point. We agree that BSP performance with predicted structures was not verified in the original manuscript. We have now conducted this evaluation using predicted unbound structures as BSP inputs. Structural graphs were constructed using the same pipeline as BAR (Appendix A.3), and CoBALT was re-evaluated under the random five-fold split. The results are tabulated in Table R1.
Table R1. BSP performance with original and predicted unbound structures under random five-fold cross-validation.
| Structural input | E-AUPRC (%) ↑ | E-AUROC (%) ↑ | E-F1 (%) ↑ | P-AUPRC (%) ↑ | P-AUROC (%) ↑ | P-F1 (%) ↑ |
|---|---|---|---|---|---|---|
| Original | 65.60 ± 1.24 | 92.71 ± 0.83 | 57.00 ± 1.75 | 88.58 ± 0.58 | 91.32 ± 0.38 | 80.73 ± 0.75 |
| Predicted unbound | 64.93 ± 1.63 | 92.65 ± 0.60 | 56.07 ± 2.08 | 88.29 ± 0.55 | 91.13 ± 0.45 | 80.38 ± 0.50 |
Results are reported as mean ± standard deviation across five folds.
All six BSP metrics remain close to the original results, with numerical decreases of less than one percentage point. Epitope and paratope AUROC are nearly unchanged, decreasing by only 0.06 and 0.19 percentage points, respectively. These results directly verify BSP performance with predicted structures and indicate that CoBALT retains robust binding-site prediction performance without bound structural inputs.
Response for Question2
Q2: Please report the key ablations across all five folds.
We thank the reviewer for this suggestion. We have repeated the complete ablation study on all five random folds and under the existing 75% and 80% antigen-homology-aware splits. For each task, all variants use identical data partitions and training configurations.
The five-fold results are reported as mean ± sample standard deviation in Table R2, and the homology-aware results are presented in Tabel R3. Overall, the multi-fold experiments reproduce the principal ablation trends. CoBALT improves all five BAR ranking metrics in 5/5 folds relative to Single BAR and improves epitope/paratope AUPRC and F1 in 5/5 folds relative to Single BSP. Removing BAR→C causes the largest average BAR degradation, while removing C, E, or the interface evidence also reduces BAR ranking performance and BSP AUPRC. Under both homology-aware splits, CoBALT achieves the highest BAR Spearman and epitope/paratope AUPRC and outperforms the single-task counterparts on all twelve metrics.
We note that w/o BAR→C achieves a slightly higher mean epitope AUROC on the random folds (93.57 versus 92.71), w/o interface in BAR achieves marginally higher P-F1 at 75% homology (75.33 versus 75.27), and w/o E achieves lower MAE (0.872 versus 0.948). However, these differences are small and inconsistent. CoBALT therefore remains competitive. For completeness, we report the full fold-level results for all variants and twelve metrics in the appendix.
Table R2. Ablation results under random five-fold cross-validation
| Model / Variant | E-AUPRC ↑ | E-AUROC ↑ | E-F1 ↑ | P-AUPRC ↑ | P-AUROC ↑ | P-F1 ↑ | MAE ↓ | Spearman ↑ | Kendall ↑ | P@1 ↑ | Hit@3 ↑ | PairAcc ↑ |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CoBALT | 65.60 ± 1.24 | 92.71 ± 0.83 | 57.00 ± 1.75 | 88.58 ± 0.58 | 91.32 ± 0.38 | 80.73 ± 0.75 | 0.65 ± 0.02 | 0.70 ± 0.01 | 0.61 ± 0.01 | 56.99 ± 1.13 | 78.36 ± 1.30 | 84.42 ± 0.36 |
| CoBALT-Single | 54.32 ± 1.78 | 91.91 ± 0.44 | 45.63 ± 3.02 | 86.78 ± 0.44 | 90.33 ± 0.29 | 78.33 ± 0.39 | 0.67 ± 0.02 | 0.66 ± 0.02 | 0.57 ± 0.02 | 53.44 ± 2.24 | 73.99 ± 3.09 | 82.23 ± 1.19 |
| w/o BAR→C | 62.63 ± 2.29 | 93.57 ± 0.67 | 47.52 ± 3.37 | 87.85 ± 0.94 | 90.98 ± 0.65 | 80.26 ± 0.76 | 0.84 ± 0.02 | 0.54 ± 0.00 | 0.47 ± 0.00 | 41.07 ± 1.58 | 61.44 ± 1.90 | 76.21 ± 0.25 |
| w/o C in BAR | 55.70 ± 2.83 | 92.06 ± 0.75 | 40.92 ± 4.43 | 83.96 ± 1.15 | 88.12 ± 0.82 | 77.27 ± 0.78 | 0.71 ± 0.03 | 0.64 ± 0.01 | 0.55 ± 0.01 | 50.59 ± 1.19 | 71.57 ± 1.83 | 81.15 ± 0.59 |
| w/o E | 55.61 ± 2.87 | 92.11 ± 0.70 | 42.37 ± 3.31 | 83.59 ± 0.63 | 87.76 ± 0.46 | 76.90 ± 0.48 | 0.72 ± 0.03 | 0.64 ± 0.01 | 0.55 ± 0.00 | 49.99 ± 1.37 | 72.09 ± 1.33 | 80.92 ± 0.32 |
| w/o interface in BAR | 55.91 ± 1.99 | 92.09 ± 0.48 | 40.60 ± 2.05 | 84.30 ± 0.88 | 88.37 ± 0.63 | 77.58 ± 0.58 | 0.72 ± 0.03 | 0.63 ± 0.00 | 0.54 ± 0.00 | 49.12 ± 2.20 | 70.79 ± 0.90 | 80.64 ± 0.16 |
Table R3. Ablation results under the 75% and 80% antigen-homology-aware splits
| Split | Model / Variant | E-AUPRC ↑ | E-AUROC ↑ | E-F1 ↑ | P-AUPRC ↑ | P-AUROC ↑ | P-F1 ↑ | MAE ↓ | Spearman ↑ | Kendall ↑ | P@1 ↑ | Hit@3 ↑ | PairAcc ↑ |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 75% | CoBALT | 53.50 | 89.38 | 49.66 | 82.30 | 87.89 | 75.27 | 0.948 | 0.368 | 0.302 | 31.76 | 64.89 | 68.76 |
| CoBALT-Single | 44.65 | 87.98 | 30.67 | 81.27 | 87.28 | 74.71 | 0.950 | 0.304 | 0.238 | 28.43 | 61.88 | 64.95 | |
| w/o BAR→C | 37.54 | 84.17 | 34.55 | 78.30 | 85.41 | 72.90 | 1.032 | 0.273 | 0.212 | 23.18 | 54.23 | 63.24 | |
| w/o C in BAR | 49.59 | 87.92 | 43.20 | 79.28 | 85.99 | 73.70 | 0.904 | 0.329 | 0.267 | 26.18 | 62.57 | 66.50 | |
| w/o E | 48.87 | 88.60 | 40.05 | 78.52 | 85.13 | 72.86 | 0.872 | 0.340 | 0.278 | 28.43 | 61.41 | 67.14 | |
| w/o interface in BAR | 44.36 | 86.51 | 40.67 | 80.73 | 87.28 | 75.33 | 0.898 | 0.365 | 0.295 | 30.04 | 63.73 | 68.72 | |
| 80% | CoBALT | 58.35 | 90.29 | 51.88 | 84.65 | 89.64 | 77.69 | 0.783 | 0.499 | 0.398 | 32.05 | 57.75 | 75.38 |
| CoBALT-Single | 46.05 | 87.84 | 30.79 | 82.93 | 88.32 | 76.31 | 0.869 | 0.467 | 0.371 | 27.68 | 55.14 | 73.61 | |
| w/o BAR→C | 51.40 | 89.77 | 44.26 | 79.75 | 86.62 | 74.52 | 0.844 | 0.396 | 0.310 | 26.23 | 54.73 | 70.17 | |
| w/o C in BAR | 43.59 | 85.83 | 31.98 | 75.71 | 82.15 | 69.27 | 0.848 | 0.446 | 0.348 | 27.28 | 52.54 | 72.71 | |
| w/o E | 37.19 | 85.03 | 32.47 | 74.04 | 81.91 | 68.97 | 0.843 | 0.446 | 0.351 | 24.90 | 50.34 | 72.86 | |
| w/o interface in BAR | 34.29 | 85.07 | 32.65 | 76.66 | 82.93 | 70.71 | 0.913 | 0.437 | 0.344 | 24.37 | 51.85 | 72.11 |
Response for Reviewer 3 (4VVh)
We sincerely thank the reviewer for the time and care devoted to evaluating our work. We appreciate your constructive feedback, which has prompted us to take a closer look at both the evaluation and presentation of the study. In the responses below, we address each point in detail and explain the revisions made to the manuscript.