ATMOS pairs a state space model with diffusion to generate atom-level molecular dynamics trajectories for proteins and protein-ligand complexes
- A generative model uses a Pairformer-based state space backbone to carry motion history forward, then a diffusion decoder produces each trajectory frame in sequence.
- It targets the gap between costly long-timescale simulation and generators that ignore temporal correlations.
- Evaluation covers protein monomers and protein-ligand systems; the authors report leading performance, though the abstract describes no prospective experimental test.
💡 Points Based on the abstract
Generating frames autoregressively is sensible, since independent-frame generators discard kinetics. But MD surrogates should be judged on slow-state populations and transition rates against long reference runs, not per-frame fidelity. For protein-ligand systems, check whether unbinding or pose switching appears at all, and whether long rollouts drift. Without code or held-out systems, this is promising rather than usable.
Atomic Trajectory Modeling with State Space Models for Biomolecular Dynamics ↗