分子之心QuantaMind登Science Advances，让AI给分子世界”拍电影”

TL;DR - QuantaMind is a reactive machine-learning force-field platform that simulates bond changes and molecular dynamics in large biological systems at near-DFT accuracy. It could move protein and drug design beyond static structure prediction toward computational analysis of reaction mechanisms.

• The published tests reached 24,001 atoms and 20 nanoseconds; post-publication extensions reportedly reached 100,000 atoms and 100 nanoseconds.
• In a 17,792-atom PETase system, QuantaMind simulated a complete catalytic cycle, with force components from 342 sampled configurations correlating above 0.99 with independent DFT calculations.
• Additional validations reproduced proton diffusion, water autoionization, and a protein-residue pKa close to experimental measurements.
• Reported applications include mechanism-guided enzyme engineering and pH-sensitive antibody design, with planned integration into the MoleculeOS platform.