ProteinMPNN-ddG
An unsupervised deep learning model for rapid and accurate prediction of protein stability changes upon mutation, based on an improved ProteinMPNN methodology.
ProteinMPNN-ddG is an unsupervised deep learning model that substantially improves protein stability prediction without requiring additional training or experimental data. It enhances the baseline ProteinMPNN model by using the full sequence context for each residue prediction and introducing a novel correction term. This term, derived from the model's prediction on a single residue's backbone atoms, nullifies background effects from amino acid abundance and geometry, making the score a better correlate for stability changes (ΔΔG) upon mutation. The method uses an efficient tied decoding scheme, enabling proteome-scale predictions at thousands of residues per second.
Estimated resource cost: 0.000527
Categories: Protein Design
Tags: Affinity Prediction Antibodies Developability Mutational Scanning Other Binders Peptides Proteins
Key Capabilities
- Predicts the effects of point mutations on protein stability (ΔΔG) using an unsupervised method.
- Improves the absolute success rate of identifying stabilizing mutations by 11 percentage points over baseline ProteinMPNN, from 66 to 70 percent on the Tsuboyama dataset.
- Uses a novel scoring method that subtracts a context-free prediction to correct for biases from amino acid frequencies and backbone geometry.
- Achieves extremely high throughput, making proteome-scale analysis feasible.
- Employs a tied decoding algorithm to efficiently compute predictions with full sequence context, reducing slowdown from N-fold to less than 4-fold for large proteins.
- Requires no additional model training or experimental stability data.
Runtime Statistics
| Metric | Value |
|---|---|
| runtime_mean | 9 |
| runtime_median | 10 |
| runtime_std | 4 |
| runtime_90th_percentile | 15 |
| runtime_max | 39 |
Similar Tools
- BindFilter
- StaB-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.