Ensemble tests mask missing dynamics in protein conformational generators
Protein generators are increasingly used to produce conformational ensembles and trajectories as faster alternatives to molecular dynamics (MD). Yet whether agreement with an equilibrium ensemble demonstrates that a generator has learned dynamics remains unresolved. Here we show that this evidence survives the destruction of dynamics. Randomly reordering authentic MD frames leaves ensemble…
Protein generators are becoming popular for creating conformational ensembles and trajectories as quicker substitutes for molecular dynamics (MD). However, it is still uncertain whether matching an equilibrium ensemble indicates that a generator has grasped dynamics. The study reveals that this evidence remains intact even after dynamics are disrupted.
Simply shuffling genuine MD frames does not affect ensemble accuracy, but significantly lowers the geometrically gated kinetic pass rate from 0.963 to 0.004. A separate MD duplicate achieves 0.95 under the same scrutiny. Dynbench distinguishes the recovery of conformational states, time-dependent behavior, and trajectory suitability.
While public generators may appear alike when subjected to ensemble tests, they diverge substantially. In a prospective assessment involving 36 proteins, the reference and benchmark remain constant, while two models swap positions. Thus, ensemble tests may artificially inflate the evidence of learned dynamics, highlighting the need for time-resolved validation as a separate criterion for assertions regarding protein dynamics generators.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
