Simulations of Protein Structure and Dynamics Beyond AlphaFold : From Structure to Function
WIREs Computational Molecular Science · 2026-11 · 16 (6)
DOI: 10.1002/wcms.70091 ↗Kurz gefasst
Fachartikel in der Zeitschrift WIREs Computational Molecular Science, online erschienen am 17.11.2026. Verfasst von Bhumika Singh und 5 weiteren Autor:innen. Die Arbeit nennt 307 Literaturangaben. Gefördert von: National Science Foundation, National Institute of General Medical Sciences.
Abstract (Originalsprache)
Artificial intelligence has changed the division of labor among computational methods used to understand biological function. For many well‐folded proteins, generating a plausible structure is no longer the principal bottleneck, and the prediction of molecular complexes continues to improve. Yet structure is not function. Current AI models generally do not determine which states are populated under specific conditions, how those states inter‐convert, or how perturbations propagate through allosteric networks, reveal cryptic pockets, or produce collective and cellular behavior. Physics‐based simulations provide a statistical‐mechanical framework for testing structural hypotheses and connecting experimental observations to conformational populations, thermodynamics, kinetics, and molecular mechanisms, although their conclusions remain limited by force‐field accuracy and sampling. We examine how AI, molecular simulation, and experiment can be combined in the Post‐AlphaFold era to move from abundant structural hypotheses toward predictive explanations of biological function. This convergence also creates an urgent need to make simulation data reusable. The Protein Data Bank transformed structural biology by establishing a shared representation that could be used far beyond the experiments that produced it. Molecular simulation data are richer but more heterogeneous, and preserving trajectories alone is insufficient. A community Molecular Dynamics Database (MDDB) could make simulations available at scale for training and evaluating new AI models. But data access alone will not produce the analogue of a PDB structure for dynamics: neither a raw trajectory nor an AI‐generated ensemble is yet a portable, interpretable object that researchers across different fields can readily use. Defining that object—one that preserves relevant states, populations, transitions, conditions, and uncertainty—remains a central challenge on the path from structures to function.
Förderung
- National Science Foundation
- National Institute of General Medical Sciences
Zitieren
APA
Singh, B., Alvarez, G., Gaza, J., Rioual, E., Sugita, Y., & Perez, A. (2026). Simulations of Protein Structure and Dynamics Beyond AlphaFold : From Structure to Function. WIREs Computational Molecular Science, 16(6). https://doi.org/10.1002/wcms.70091
Vancouver
Singh B, Alvarez G, Gaza J, Rioual E, Sugita Y, Perez A. Simulations of Protein Structure and Dynamics Beyond AlphaFold : From Structure to Function. WIREs Computational Molecular Science. 2026;16(6). doi:10.1002/wcms.70091
BibTeX
@article{Singh2026,
author = {Singh, Bhumika and Alvarez, Guadalupe and Gaza, Jokent and Rioual, Elisa and Sugita, Yuji and Perez, Alberto},
title = {Simulations of Protein Structure and Dynamics Beyond AlphaFold : From Structure to Function},
journal = {WIREs Computational Molecular Science},
year = {2026},
volume = {16},
number = {6},
doi = {10.1002/wcms.70091}
}
EUVIMED