StrucNS reveals interaction-weighted network topology as the driving predictor of absolute stability of natural and de novo proteins.

Interaction-weighted network topology predicts absolute stability of natural and de novo proteins with high accuracy, outperforming existing models on mutational stability tests.

• Why it matters: Understanding how protein structure and sequence influence stability is crucial for designing functional proteins, yet current methods struggle to decode the complex physicochemical interactions involved.
• What they did: StrucNS employs network science to analyze protein folds as interaction-weighted networks, extracting stability-related signals directly from topological features without relying on high-dimensional evolutionary data, tested on 3 miniprotein sets and de novo designs.
• The result: The framework surpasses unsupervised models and matches or exceeds supervised approaches in predicting stability, with interpretability analyses revealing key topological features like connectivity between hydrophobic core and surface as critical stability determinants.