StaB-ddG
Fast & accurate deep learning model for predicting binding ∆∆G using folding energy principles and a ProteinMPNN-based inverse folding framework.
StaB-ddG is a deep learning model for predicting mutational effects on protein–protein binding affinity, achieving performance on par with state-of-the-art force field methods while being over 1000× faster. Built on a thermodynamic identity linking binding free energy to folding stability, STAB-DDG leverages a zero-shot inverse folding model (ProteinMPNN) and is fine-tuned on high-throughput folding and binding ∆∆G datasets. It incorporates variance reduction techniques and satisfies key physical constraints (antisymmetry, path-independence), offering robust performance across benchmark datasets. The model supports multi-chain complexes and multiple mutations, delivering accurate predictions for binding energy changes relevant to structural biology and therapeutic design.
Estimated resource cost: 0.000527
Categories: Protein Design
Tags: Affinity Prediction Antibodies Developability Mutational Scanning Other Binders Peptides Proteins
Key Capabilities
- Thermodynamic parameterization defines binding ∆∆G as a difference in folding free energies.
- Initialized from ProteinMPNN inverse folding model for strong zero-shot folding stability predictions.
- Supports multi-chain complexes and multiple mutations without structural heuristics.
- Fine-tuned sequentially on large folding and smaller binding ∆∆G datasets for improved generalization.
- Satisfies antisymmetry and mutational path-independence by design, enabling physically grounded predictions.
- Variance reduction via Monte Carlo ensembling lowers inference noise.
- Outperforms prior DL predictors and matches FoldX accuracy with 1000X speedup.
- Validates across SKEMPIv2.0, de novo binder assays, and TCR-mimic antibody datasets.
Runtime Statistics
| Metric | Value |
|---|---|
| runtime_mean | 67 |
| runtime_median | 16 |
| runtime_std | 187 |
| runtime_90th_percentile | 109 |
| runtime_max | 1592 |
Similar Tools
- BindFilter
- ProteinMPNN-ddG
- BioBind
- BioPhi
Ready to submit your job?
Review your configuration, then confirm the estimated credit cost before you run the job. Note that credit estimates are not guaranteed and runtime can vary depending on inputs and settings.
Estimated Credits: 0.000527
Invite-only, limited-time access. Please contact ztang@getantibody.com.
Invite-only, limited-time access. Please contact ztang@getantibody.com.