Pegado, F. J. d. O.; Ortega, J. M.; Silva, J. R. P.

We developed AlphaUnfold, an automated pipeline that couples AF3 predictions with short-time (5 ns) high-pressure Molecular Dynamics (MD) using NAMD3. By subjecting models to baric stress, AlphaUnfold acts as a dynamic stress-test to identify structural fragility and potential unfolding. Testing a diverse set of proteins revealed a significant inverse correlation between average pLDDT and Root Mean Square Deviation (RMSD) after MD, indicating that lower confidence translates to rapid structural drift. Furthermore, domains with low local pLDDT consistently exhibited high Root Mean Square Fluctuation (RMSF), a behavior also observed in 200 ns simulations under standard pressure, pinpointing specific metastable areas. AlphaUnfold thus provides a viable, computationally efficient framework for assessing the biophysical robustness of AI-generated models, offering an experimental-like validation that ensures more reliable downstream applications in structural biology.