Diffusion-Model-Guided Adaptive Sampling of Rare Conformational States in Adenylate Kinase
- 1Center for Management of Information Technologies, Kagoshima University, 1-21-35 Korimoto, Kagoshima, Japan, 890-0065
Abstract
Efficiently exploring the rare conformational states of biomolecules remains a central challenge in molecular dynamics simulation. Here we show, using adenylate kinase as a methodological testbed, that the static structural scores conventionally used to prioritize diffusion-modelgenerated candidates have essentially no power to predict which candidates actually reach a kinetically rare state upon subsequent simulation (AUC = 0.500). Motivated by this finding, we introduce two adaptive strategies — direct PCA-distance targeting and replication of confirmed-hit structures — that achieve significantly higher rare-state hit rates and dwell times than extended conventional molecular dynamics or random-initialization sampling under matched computational budgets, while also enabling, for the first time, statistically stable estimation of rare-macrostate populations. These results show that structural novelty and kinetically meaningful entry into a rare state are not the same property, providing concrete design guidance for generative-model-driven adaptive sampling in other biomolecular systems.
Related
- Comparative Conformational Sampling of Apo Adenylate Kinase by Enhanced Molecular Dynamics and a Sequence-Based Generative Ensemble Model
- Achieving Sustained Conformational Exploration with Continually Adaptive Local Memory Sampling
- Breaking timescales with generative sampling of conformational transitions
- Shedding light on Gaussia Luciferase conformational space using multi-conformation enhanced sampling simulation
- Conformational Sampling of CPT1A Enables Dynamic Structure-Based Ligand Screening
BibTeX
@misc{sakae2026diffusionmodelguided,
title = {{Diffusion-Model-Guided Adaptive Sampling of Rare Conformational States in Adenylate Kinase}},
author = {Yoshitake Sakae},
howpublished = {ChemRxiv},
year = {2026},
month = sep,
doi = {10.26434/chemrxiv.15008478/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008478/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Diffusion-Model-Guided Adaptive Sampling of Rare Conformational States in Adenylate Kinase
AU - Yoshitake Sakae
PY - 2026
DA - 2026/09/08
PB - ChemRxiv
DO - 10.26434/chemrxiv.15008478/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008478/v1
AB - Efficiently exploring the rare conformational states of biomolecules remains a central challenge in molecular dynamics simulation. Here we show, using adenylate kinase as a methodological testbed, that the static structural scores conventionally used to prioritize diffusion-modelgenerated candidates have essentially no power to predict which candidates actually reach a kinetically rare state upon subsequent simulation (AUC = 0.500). Motivated by this finding, we introduce two adaptive strategies — direct PCA-distance targeting and replication of confirmed-hit structures — that achieve significantly higher rare-state hit rates and dwell times than extended conventional molecular dynamics or random-initialization sampling under matched computational budgets, while also enabling, for the first time, statistically stable estimation of rare-macrostate populations. These results show that structural novelty and kinetically meaningful entry into a rare state are not the same property, providing concrete design guidance for generative-model-driven adaptive sampling in other biomolecular systems.
ER -