FamGeo-DDG: mutation-centered local graphs plus ESM-2 for PPI ΔΔG prediction
- Predicts how a mutation shifts protein–protein binding energy using a local residue graph built around the mutated site.
- Node features combine the residue substitution, DSSP secondary-structure descriptors and ESM-2 embedding differences.
- Retrospective benchmarking on several public mutation datasets under three cross-validation split schemes, including a family-level split, against 13 baseline methods; ablations and an external validation set are included.
- The authors report competitive performance on the family-level split; no prospective testing is described.
💡 Points Based on the abstract
The family-level split is the only one of the three schemes that says anything about novel targets; random splits leak near-identical complexes across folds. Even so, PPI ΔΔG sets are small, redundant and skewed toward destabilizing alanine mutations, so a gain over 13 baselines may not survive contact with FEP or empirical tools on a real affinity-maturation campaign. Prospective designs are the missing test.
📄 Paper · Journal of Chemical Information and Modeling ↗ FamGeo-DDG: A Mutation-Centered Local Geometric-Semantic Framework for Protein–Protein ΔΔG Prediction