Research map / Probabilistic and scientific modeling / Molecules and proteins
A Hierarchical Training Paradigm for Antibody Structure-sequence Co-design
Fang Wu, Stan Z. Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add single-protein sequence level (ESM-2) training (ablation) | HTP baseline · SAbDab CDR-H3 co-design | AAR (%): 22.95 -> 33.87 | Improved |
| add single-protein sequence level (ESM-2) training (ablation) | HTP baseline · SAbDab CDR-H3 co-design | RMSD: 3.55 -> 2.77 | Improved |
| add antibody sequence level training (ablation) | HTP with SPS · SAbDab CDR-H3 co-design | AAR (%): 33.87 -> 38.42 | Improved |
| add protein-protein complex structure level training (ablation) | HTP with SPS and AS · SAbDab CDR-H3 co-design | RMSD: 2.49 -> 2.06 | Improved |
| add protein-protein complex structure level training (ablation) | HTP with SPS and AS · SAbDab CDR-H3 co-design | AAR (%): 38.42 -> 40.98 | Improved |
| replace pretrained PLM features with learnable embeddings (ablation) | HTP · SAbDab CDR-H3 co-design | AAR (%): 40.98 -> 25.31 | Hurt |
| replace pretrained PLM features with learnable embeddings (ablation) | HTP · SAbDab CDR-H3 co-design | RMSD: 2.06 -> 2.65 | Hurt |
| use external AbLang instead of HTP's first two pretraining levels (ablation) | HTP · SAbDab CDR-H3 co-design | AAR (%): 28.46 vs 40.98 | Hurt |
| use external AntiBERTa instead of HTP's first two pretraining levels (ablation) | HTP · SAbDab CDR-H3 co-design | AAR (%): 34.52 vs 40.98 | Hurt |
| add AntiBERTa features to GNN without any PLM (ablation) | GNN baseline · SAbDab CDR-H3 co-design | AAR (%): 25.31 -> 34.52 | Improved |
| add AbLang features to GNN without any PLM (ablation) | GNN baseline · SAbDab CDR-H3 co-design | AAR (%): 25.31 -> 28.46 | Improved |
| use sequence-only ESM-2 with MLP for CDR-H3 recovery (ablation) | ESM-2 only · SAbDab CDR-H3 co-design | AAR (%): 14.63 vs 40.98 | Hurt |
Closest papers by meaning
- Cross-Gate MLP with Protein Complex Invariant Embedding Is a One-Shot Antibody Designer (AAAI 2024 · similarity 0.91)
- Relation-Aware Equivariant Graph Networks for Epitope-Unknown Antibody Design and Specificity Optimization (AAAI 2025 · similarity 0.90)
- Pareto-Optimal Energy Alignment for Designing Nature-Like Antibodies (NeurIPS 2025 · similarity 0.90)
- Antigen-Specific Antibody Design via Direct Energy-based Preference Optimization (NeurIPS 2024 · similarity 0.89)
- Synergy of GFlowNet and Protein Language Model Makes a Diverse Antibody Designer (AAAI 2025 · similarity 0.89)
- Antibody Design Using a Score-based Diffusion Model Guided by Evolutionary, Physical and Geometric Constraints (ICML 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
