Methods for antibody-antigen docking. The dataset part shows the number of protein or Ab–Ag complexes (complx). To compare designs, we consider AAR and RMSD between the original and generated sequences and structures. For docking evaluation, we utilize DockQ[157] and success rate (SSR) to compare the original docked complex with the predicted one. In this table, Transformers are abbreviated as TF. *BC40 is available at https://drug.ai.tencent.com/protein/bc40/download.html.
| Name . | Class . | Model . | Training dataset . | Design . | Docking . | Description . | Ref . |
|---|---|---|---|---|---|---|---|
| DLAB | Antibody | CNN | SAbDab (1.2k Ab–Ag complx.) | - | - | Strengths: Improved pose-ranking. Limitations: Use of rigid docking instead of flexible docking. Applications: Early-stage virtual screening of Ab (known Ag). | [145] |
| DockGPT | Antibody | TF | BC40* 37k chains DIPS [146] (33k complx.) SAbDab (2.4k Ab–Ag complx.) | RMSD H1: 1.11 Å H2: 1.02 Å H3: 1.88 Å | DockQ: 26.1% | Strengths: Circumvents explicit training on bound structures and offers a natural approach to modeling conformational flexibility in complex prediction. Limitations: use only single atom type and threshold to provide the model with interface and contact information. Applications: Flexible and site-specific protein docking; Dock and CDR design for a specific epitope. | [147] |
| dyMEAN | Antibody | MEAN | SAbDab Design: 3k Ab Docking: 60 Ab–Ag complx. | AAR Full: 74.96% CDRs: 60.07% H3: 43.65% | DockQ Full: 41.2% CDRs: 39.6% H3: 40.9% | Strengths: Multi-channel encoder addresses the issue of varying numbers of atoms in different residues in full-atom modeling. Limitations: Cannot design rational antibodies [148]. Applications: CDR design and docking considering the epitope structure and Ab incomplete sequence. | [149] |
| GeoDock | Protein | TF | DIPS [146] (36k complx.) DB5.5 [150] (178 complx.) | - | SSR: 41% | Strengths: Uses sequence and structure embeddings. Limitations: Does not outperforms methods that use sampling and re-ranking. Applications: Flexible protein-protein docking. | [151] |
| HERN | Antibody | GNN | SAbDab (3k Ab–Ag complx.) | H3 AAR: 34.1% | Paratope DockQ: 43.8 % SSR: 100% | Strengths: Represents binding interface as a dynamic hierarchical graph. Limitations: Needs to be combined with epitope prediction approaches, focus only on CDR-H3. Applications: Paratope docking and design given the epitope. | [152] |
| Peng et al. | Antibody | AbDesign: MC-EGNN AbDock: IPA | SAbDab | RMSD: 2.56 Å AAR: 36.47% | DockQ H chain: 26% CDRH: 30% H3: 44% | Strengths: Integrates generative diffusion models for diverse candidate sampling. Limitations: Depends on the presence of Ab–Ag complex structures for optimization. Applications: CDR design and docking to improve binding affinity. | [153] |
| PointDE | Protein Antibody | PMLP | DOCK- GROUND [154] (61 complx.) IEDB [155] (659 Ab–Ag complx.) | - | SSR proteins: 65.6% Ab–Ag: 56.6% | Strengths: First to use point cloud for protein docking evaluation. Limitations: Uses just PDB information. Applications: Docking evaluation. | [156] |
| Name . | Class . | Model . | Training dataset . | Design . | Docking . | Description . | Ref . |
|---|---|---|---|---|---|---|---|
| DLAB | Antibody | CNN | SAbDab (1.2k Ab–Ag complx.) | - | - | Strengths: Improved pose-ranking. Limitations: Use of rigid docking instead of flexible docking. Applications: Early-stage virtual screening of Ab (known Ag). | [145] |
| DockGPT | Antibody | TF | BC40* 37k chains DIPS [146] (33k complx.) SAbDab (2.4k Ab–Ag complx.) | RMSD H1: 1.11 Å H2: 1.02 Å H3: 1.88 Å | DockQ: 26.1% | Strengths: Circumvents explicit training on bound structures and offers a natural approach to modeling conformational flexibility in complex prediction. Limitations: use only single atom type and threshold to provide the model with interface and contact information. Applications: Flexible and site-specific protein docking; Dock and CDR design for a specific epitope. | [147] |
| dyMEAN | Antibody | MEAN | SAbDab Design: 3k Ab Docking: 60 Ab–Ag complx. | AAR Full: 74.96% CDRs: 60.07% H3: 43.65% | DockQ Full: 41.2% CDRs: 39.6% H3: 40.9% | Strengths: Multi-channel encoder addresses the issue of varying numbers of atoms in different residues in full-atom modeling. Limitations: Cannot design rational antibodies [148]. Applications: CDR design and docking considering the epitope structure and Ab incomplete sequence. | [149] |
| GeoDock | Protein | TF | DIPS [146] (36k complx.) DB5.5 [150] (178 complx.) | - | SSR: 41% | Strengths: Uses sequence and structure embeddings. Limitations: Does not outperforms methods that use sampling and re-ranking. Applications: Flexible protein-protein docking. | [151] |
| HERN | Antibody | GNN | SAbDab (3k Ab–Ag complx.) | H3 AAR: 34.1% | Paratope DockQ: 43.8 % SSR: 100% | Strengths: Represents binding interface as a dynamic hierarchical graph. Limitations: Needs to be combined with epitope prediction approaches, focus only on CDR-H3. Applications: Paratope docking and design given the epitope. | [152] |
| Peng et al. | Antibody | AbDesign: MC-EGNN AbDock: IPA | SAbDab | RMSD: 2.56 Å AAR: 36.47% | DockQ H chain: 26% CDRH: 30% H3: 44% | Strengths: Integrates generative diffusion models for diverse candidate sampling. Limitations: Depends on the presence of Ab–Ag complex structures for optimization. Applications: CDR design and docking to improve binding affinity. | [153] |
| PointDE | Protein Antibody | PMLP | DOCK- GROUND [154] (61 complx.) IEDB [155] (659 Ab–Ag complx.) | - | SSR proteins: 65.6% Ab–Ag: 56.6% | Strengths: First to use point cloud for protein docking evaluation. Limitations: Uses just PDB information. Applications: Docking evaluation. | [156] |
Methods for antibody-antigen docking. The dataset part shows the number of protein or Ab–Ag complexes (complx). To compare designs, we consider AAR and RMSD between the original and generated sequences and structures. For docking evaluation, we utilize DockQ[157] and success rate (SSR) to compare the original docked complex with the predicted one. In this table, Transformers are abbreviated as TF. *BC40 is available at https://drug.ai.tencent.com/protein/bc40/download.html.
| Name . | Class . | Model . | Training dataset . | Design . | Docking . | Description . | Ref . |
|---|---|---|---|---|---|---|---|
| DLAB | Antibody | CNN | SAbDab (1.2k Ab–Ag complx.) | - | - | Strengths: Improved pose-ranking. Limitations: Use of rigid docking instead of flexible docking. Applications: Early-stage virtual screening of Ab (known Ag). | [145] |
| DockGPT | Antibody | TF | BC40* 37k chains DIPS [146] (33k complx.) SAbDab (2.4k Ab–Ag complx.) | RMSD H1: 1.11 Å H2: 1.02 Å H3: 1.88 Å | DockQ: 26.1% | Strengths: Circumvents explicit training on bound structures and offers a natural approach to modeling conformational flexibility in complex prediction. Limitations: use only single atom type and threshold to provide the model with interface and contact information. Applications: Flexible and site-specific protein docking; Dock and CDR design for a specific epitope. | [147] |
| dyMEAN | Antibody | MEAN | SAbDab Design: 3k Ab Docking: 60 Ab–Ag complx. | AAR Full: 74.96% CDRs: 60.07% H3: 43.65% | DockQ Full: 41.2% CDRs: 39.6% H3: 40.9% | Strengths: Multi-channel encoder addresses the issue of varying numbers of atoms in different residues in full-atom modeling. Limitations: Cannot design rational antibodies [148]. Applications: CDR design and docking considering the epitope structure and Ab incomplete sequence. | [149] |
| GeoDock | Protein | TF | DIPS [146] (36k complx.) DB5.5 [150] (178 complx.) | - | SSR: 41% | Strengths: Uses sequence and structure embeddings. Limitations: Does not outperforms methods that use sampling and re-ranking. Applications: Flexible protein-protein docking. | [151] |
| HERN | Antibody | GNN | SAbDab (3k Ab–Ag complx.) | H3 AAR: 34.1% | Paratope DockQ: 43.8 % SSR: 100% | Strengths: Represents binding interface as a dynamic hierarchical graph. Limitations: Needs to be combined with epitope prediction approaches, focus only on CDR-H3. Applications: Paratope docking and design given the epitope. | [152] |
| Peng et al. | Antibody | AbDesign: MC-EGNN AbDock: IPA | SAbDab | RMSD: 2.56 Å AAR: 36.47% | DockQ H chain: 26% CDRH: 30% H3: 44% | Strengths: Integrates generative diffusion models for diverse candidate sampling. Limitations: Depends on the presence of Ab–Ag complex structures for optimization. Applications: CDR design and docking to improve binding affinity. | [153] |
| PointDE | Protein Antibody | PMLP | DOCK- GROUND [154] (61 complx.) IEDB [155] (659 Ab–Ag complx.) | - | SSR proteins: 65.6% Ab–Ag: 56.6% | Strengths: First to use point cloud for protein docking evaluation. Limitations: Uses just PDB information. Applications: Docking evaluation. | [156] |
| Name . | Class . | Model . | Training dataset . | Design . | Docking . | Description . | Ref . |
|---|---|---|---|---|---|---|---|
| DLAB | Antibody | CNN | SAbDab (1.2k Ab–Ag complx.) | - | - | Strengths: Improved pose-ranking. Limitations: Use of rigid docking instead of flexible docking. Applications: Early-stage virtual screening of Ab (known Ag). | [145] |
| DockGPT | Antibody | TF | BC40* 37k chains DIPS [146] (33k complx.) SAbDab (2.4k Ab–Ag complx.) | RMSD H1: 1.11 Å H2: 1.02 Å H3: 1.88 Å | DockQ: 26.1% | Strengths: Circumvents explicit training on bound structures and offers a natural approach to modeling conformational flexibility in complex prediction. Limitations: use only single atom type and threshold to provide the model with interface and contact information. Applications: Flexible and site-specific protein docking; Dock and CDR design for a specific epitope. | [147] |
| dyMEAN | Antibody | MEAN | SAbDab Design: 3k Ab Docking: 60 Ab–Ag complx. | AAR Full: 74.96% CDRs: 60.07% H3: 43.65% | DockQ Full: 41.2% CDRs: 39.6% H3: 40.9% | Strengths: Multi-channel encoder addresses the issue of varying numbers of atoms in different residues in full-atom modeling. Limitations: Cannot design rational antibodies [148]. Applications: CDR design and docking considering the epitope structure and Ab incomplete sequence. | [149] |
| GeoDock | Protein | TF | DIPS [146] (36k complx.) DB5.5 [150] (178 complx.) | - | SSR: 41% | Strengths: Uses sequence and structure embeddings. Limitations: Does not outperforms methods that use sampling and re-ranking. Applications: Flexible protein-protein docking. | [151] |
| HERN | Antibody | GNN | SAbDab (3k Ab–Ag complx.) | H3 AAR: 34.1% | Paratope DockQ: 43.8 % SSR: 100% | Strengths: Represents binding interface as a dynamic hierarchical graph. Limitations: Needs to be combined with epitope prediction approaches, focus only on CDR-H3. Applications: Paratope docking and design given the epitope. | [152] |
| Peng et al. | Antibody | AbDesign: MC-EGNN AbDock: IPA | SAbDab | RMSD: 2.56 Å AAR: 36.47% | DockQ H chain: 26% CDRH: 30% H3: 44% | Strengths: Integrates generative diffusion models for diverse candidate sampling. Limitations: Depends on the presence of Ab–Ag complex structures for optimization. Applications: CDR design and docking to improve binding affinity. | [153] |
| PointDE | Protein Antibody | PMLP | DOCK- GROUND [154] (61 complx.) IEDB [155] (659 Ab–Ag complx.) | - | SSR proteins: 65.6% Ab–Ag: 56.6% | Strengths: First to use point cloud for protein docking evaluation. Limitations: Uses just PDB information. Applications: Docking evaluation. | [156] |
This PDF is available to Subscribers Only
View Article Abstract & Purchase OptionsFor full access to this pdf, sign in to an existing account, or purchase an annual subscription.
