AlphaDynamics: A Per-System Phase-Flow Propagator for Protein Torsion Dynamics with Calibrated Rollout Fidelity
Fuente:
Zenodo
Salvato in:
| Autore principale: | |
|---|---|
| Natura: | Recurso digital |
| Pubblicazione: |
Zenodo
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866902102694952960 |
|---|---|
| author | Gwozdz, Krzysztof |
| author_facet | Gwozdz, Krzysztof |
| contents | Reviewer-hardened v2 release of AlphaDynamics, a per-system neural propagator for protein backbone torsion dynamics. This release adds an anchored rollout-fidelity audit, AR(1) baseline, statistical tests, K and kappa ablations, and a shared 4AA peptide head-to-head against Microsoft Timewarp. With calibrated kappa x1 rollout inference, AlphaDynamics achieves mean Ramachandran JSD 0.014 versus 0.356 for the pretrained Microsoft Timewarp 4AA model on three out-of-training tetrapeptides. Scope: AlphaDynamics is trained from target-specific seed MD and is not a zero-shot sequence-to-dynamics model. |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_19998622 |
| institution | Zenodo |
| language | |
| publishDate | 2026 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | AlphaDynamics: A Per-System Phase-Flow Propagator for Protein Torsion Dynamics with Calibrated Rollout Fidelity Gwozdz, Krzysztof protein dynamics molecular dynamics torsion angles phase oscillators neural ODE mdCATH Timewarp Ramachandran rollout fidelity Reviewer-hardened v2 release of AlphaDynamics, a per-system neural propagator for protein backbone torsion dynamics. This release adds an anchored rollout-fidelity audit, AR(1) baseline, statistical tests, K and kappa ablations, and a shared 4AA peptide head-to-head against Microsoft Timewarp. With calibrated kappa x1 rollout inference, AlphaDynamics achieves mean Ramachandran JSD 0.014 versus 0.356 for the pretrained Microsoft Timewarp 4AA model on three out-of-training tetrapeptides. Scope: AlphaDynamics is trained from target-specific seed MD and is not a zero-shot sequence-to-dynamics model. |
| title | AlphaDynamics: A Per-System Phase-Flow Propagator for Protein Torsion Dynamics with Calibrated Rollout Fidelity |
| topic | protein dynamics molecular dynamics torsion angles phase oscillators neural ODE mdCATH Timewarp Ramachandran rollout fidelity |
| url | https://doi.org/10.5281/zenodo.19998622 |