This constraint is open source. Any third-party models, product names, or trademarks referenced are the property of their respective owners, and Proto is not affiliated with them.

```bibtex theme={null} @article{passaro2025boltz2, title={Boltz-2: Towards Accurate and Efficient Binding Affinity Prediction}, author={Passaro, Saro and Corso, Gabriele and Wohlwend, Jeremy and Reveiz, Mateo and Thaler, Stephan and Somnath, Vignesh Ram and Getz, Noah and Portnoi, Tally and Roy, Julien and St{\"a}rk, Hannes and Kwabi-Addo, David and Beaini, Dominique and Jaakkola, Tommi and Barzilay, Regina}, journal={bioRxiv}, year={2025}, doi={10.1101/2025.06.14.659707}, publisher={Cold Spring Harbor Laboratory} } @article{wohlwend2024boltz1, title={Boltz-1: Democratizing Biomolecular Interaction Modeling}, author={Wohlwend, Jeremy and Corso, Gabriele and Passaro, Saro and Reveiz, Mateo and Leidal, Ken and Swanson, Wojtek and Turnbull, Robert and Shuaibi, Muhammed and Ahdritz, Gustaf and Getz, Gad and Jaakkola, Tommi and Barzilay, Regina}, journal={bioRxiv}, year={2024}, doi={10.1101/2024.11.19.624167}, publisher={Cold Spring Harbor Laboratory} } ```

Copy citation

ConfigBoltzBindingStrengthConfig Source

Configuration for Boltz binding strength constraint. This class defines configuration parameters for evaluating protein-protein, protein-ligand, and protein-nucleic-acid (DNA/RNA) binding using Boltz, a biomolecular structure prediction model. A common use case is scoring a protein occupying a nucleic-acid target (e.g. a repressor bound to a DNA operator) via interface confidence. Boltz predicts complex structures and provides confidence metrics for binding quality, interface accuracy, and overall structure reliability. The constraint evaluates these metrics against target values to assess binding strength and quality. The constraint uses a penalty-based scoring system where each metric is evaluated against its target value and tolerance. Metrics are classified as "higher is better" (e.g., interface confidence scores) or "lower is better" (e.g., predicted distance errors). Penalties are combined using weighted averages, with default weights chosen by complex type (monomer, protein-ligand, or multi-chain — the latter covering both protein-protein and protein-nucleic-acid complexes). **Metric interpretation:** * **iptm/ligand\_iptm/protein\_iptm**: Interface confidence (0-1). Higher = better binding prediction. Values >0.8 indicate confident binding interfaces. * **complex\_iplddt**: Interface per-residue confidence (0-1). Higher = more reliable interface residue predictions. * **complex\_plddt**: Overall structure confidence (0-1). Similar to ESMFold pLDDT. * **ptm**: Overall structural accuracy (0-1). Similar to ESMFold pTM. * **complex\_ipde/complex\_pde**: Predicted distance errors in Ångströms. Lower = more accurate structure. Values \<3 Å indicate high accuracy. * **confidence\_score**: Boltz's aggregate confidence combining multiple factors. Target values for 'higher is better' metrics. Target values for 'lower is better' metrics. Tolerances for higher-is-better metrics; once a value falls this far below target, penalty hits 1.0. Tolerances for lower-is-better metrics (Å); once exceeding target by this much, penalty hits 1.0. Weights for combining penalties Whether to include confidence\_score in penalty calculation (adds weight 0.10) Component to return: 'total\_penalty' (weighted combination) or specific metric name Options: `total_penalty`, `iptm`, `ligand_iptm`, `protein_iptm`, `complex_iplddt`, `complex_plddt`, `complex_pde`, `complex_ipde`, `confidence_score`, `ptm` Boltz2 configuration for structure prediction.