How to Use StaB-ddG

StaB-ddG is a structure-based model for predicting how mutations change protein-protein binding free energy. It is designed for interface engineering problems where researchers need to compare variant panels, rescue weakened interactions, or identify substitutions that are likely to disrupt or strengthen binding before committing to experimental measurement.

The method is especially relevant for antibody engineering, affinity maturation, complex optimization, and mechanistic mutational analysis because it works directly from a bound complex and explicit mutation definitions rather than from sequence alone.

On Neurosnap, researchers upload an Input Structure, define the two binding partners with Interface Chain partner(s) 1 and Interface Chain partner(s) 2, and specify one or more Mutants. Number of Monte Carlo samples can be increased when a project needs a more stable estimate for borderline variants.

StaB-ddG is built around a thermodynamic view of binding in which binding ΔΔG is parameterized through folding free-energy differences. The model is initialized from ProteinMPNN, giving it a strong inverse-folding prior, and then refined on large folding- and binding-perturbation datasets. The paper also emphasizes physically motivated constraints such as antisymmetry and path independence, which matter because binding-energy predictions should remain self-consistent across forward and reverse mutation paths.

This design makes StaB-ddG different from simple heuristic interface scorers. It supports multi-chain complexes and multiple simultaneous mutations without reducing the problem to one residue at a time, and Monte Carlo ensembling is used to reduce variance in the final estimate. That combination is useful for realistic protein-engineering campaigns where candidate variants are often combinatorial rather than single-site.

On Neurosnap, the output is best interpreted comparatively across a candidate set. Researchers can rank mutations by predicted binding impact, separate clearly destabilizing edits from potentially beneficial ones, and decide which designs merit deeper structural review or experimental affinity measurement.

• Complex-aware input: StaB-ddG starts from the actual bound structure and explicit partner chains, which keeps the prediction tied to the interface under study.
• Thermodynamically grounded model: The method uses a folding-energy formulation with physical consistency constraints rather than only empirical interface heuristics.
• Multi-mutation support: Single substitutions and combinatorial mutant sets can be screened in the same workflow.
• Experiment-facing ranking: The predictions are well suited to narrowing variant panels before binding assays or deeper simulation.