DG Prediction
1 Introduction
The therapeutic efficacy of protein drugs is highly dependent on protein-protein binding (PPB) affinity. High-affinity molecules can bind targets more specifically, improving drug efficacy and reducing off-target side effects. Therefore, rapid and accurate assessment of protein interaction affinity is a core step in the early screening of protein and antibody drugs. Currently, affinity measurement mainly relies on wet-lab techniques such as protein microarrays and Octet. Although these assays are accurate, their workflows are complex and costly, making it difficult to meet the needs of high-throughput, large-scale candidate molecule screening.
The PPBind1 model uses three datasets: the large-scale protein affinity dataset PPB-Affinity1, multiple heterogeneous DMS affinity datasets, and the dimeric structure dataset DIPS-Plus. Through monotonic neural networks and an innovative "feature alignment" mechanism for multi-task learning, it constructs two models, PPBind-3D and PPBind-1D, and validates their practical value in three virtual screening cases for antibody and nanobody affinity ranking tasks.
Figure 1. PPBind model framework
Figure 2. Performance of the PPBind model in virtual screening of HEL nanobodies
2 Parameter Description
- PDB File (Required): Upload a PDB file of a protein complex.
- Receptor Chain (Required): Specify the chain ID in the PDB file that belongs to the receptor.
- Ligand Chain (Required): Specify the chain ID in the PDB file that belongs to the ligand.
- Mutation List (Optional): Enter a mutation string, e.g., "FP2Y" indicates that position 2 on chain P is mutated from F to Y. If multiple mutations are present at the same time, separate them with commas. Only one variant input is supported.
3 Result Description
| Column Name | Description |
|---|---|
| ID | Consists of the input PDB filename, the specified Receptor Chain and Ligand Chain, and the Mutation List |
| DG Pred | Affinity value predicted by the model, in kcal/mol |
| Kd | Affinity value predicted by the model, in mol/L |
4 References
[1] Peiyi Chen, Huaqing Liu, Shuxian Zhou, Lunji Wang, Mao Pang, Jing Zhao, Guoxin Fan, Lanqing Han, PPBind-3D/1D: Protein-Protein Binding Affinity Prediction for Structural and Sequence Data, Antibody Therapeutics, 2026;, tbag053, https://doi.org/10.1093/abt/tbag053.
[2] Liu, H., Chen, P., Zhai, X. et al. PPB-Affinity: Protein-Protein Binding Affinity dataset for AI-based protein drug discovery. Sci Data 11, 1316 (2024). https://doi.org/10.1038/s41597-024-03997-4.