UGME: A python package for non-Markovian protein folding dynamics
Abstract
Biomolecular processes, including protein folding, protein-protein association, and protein-DNA recognition, often occur on timescales beyond what individual atomistic molecular dynamics (MD) simulations can reach. To access these longer-timescale dynamics, practitioners commonly combine many parallelized, short trajectories into reduced kinetic models, such as Markov state models (MSMs), that describe how the biomolecules interconvert between a handful of biomolecular configurations (e.g., folded and unfolded states of a protein). Central to MSM construction is the assumption that, at sufficiently long time lags, these projected dynamics become memoryless, such that the future evolution of the system depends only on its current state. Yet, projecting high-dimensional biomolecular dynamics onto a small set of states necessarily discards information about unresolved degrees of freedom, including solvent reorganization, side-chain rearrangements, and intermediate configurations. Because these unresolved motions can continue to influence subsequent dynamics, the resulting dynamics retain memory that is absent from a Markovian description. Here, we present ugme, a modular, open-source Python package for constructing and validating generalized master equations (GME) models that explicitly account for these non-Markovian (memory) effects.
Related
- TASTET: a Python package for geometry-driven selection of representative atomistic structures from large configurational spaces
- libNLPBE: An Open-Source Python Package for Solving the Non-Linear Poisson-Boltzmann Equation
- PyCC: A Reference Quantum Chemistry Python Package for Education, Validation, and Rapid Prototyping
- PolyNet: A Python Package for Polymer Informatics
- Folding and Function as Two Sides of One Coin: Statistical Mechanics, Information Theory, and Deep Learning of Binding, Catalysis, and Allostery in an Exactly Enumerable Lattice Protein
BibTeX
@misc{dominic2026ugme,
title = {{UGME: A python package for non-Markovian protein folding dynamics}},
author = {Anthony J. Dominic III and Bertram Cham and Andrés Montoya-Castillo},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.14234},
eprint = {2610.14234},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.14234},
}
RIS
TY - UNPB
TI - UGME: A python package for non-Markovian protein folding dynamics
AU - Anthony J. Dominic III
AU - Bertram Cham
AU - Andrés Montoya-Castillo
PY - 2026
DA - 2026/10/09
PB - arXiv
DO - 10.48550/arxiv.2610.14234
UR - https://arxiv.org/abs/2610.14234
AB - Biomolecular processes, including protein folding, protein-protein association, and protein-DNA recognition, often occur on timescales beyond what individual atomistic molecular dynamics (MD) simulations can reach. To access these longer-timescale dynamics, practitioners commonly combine many parallelized, short trajectories into reduced kinetic models, such as Markov state models (MSMs), that describe how the biomolecules interconvert between a handful of biomolecular configurations (e.g., folded and unfolded states of a protein). Central to MSM construction is the assumption that, at sufficiently long time lags, these projected dynamics become memoryless, such that the future evolution of the system depends only on its current state. Yet, projecting high-dimensional biomolecular dynamics onto a small set of states necessarily discards information about unresolved degrees of freedom, including solvent reorganization, side-chain rearrangements, and intermediate configurations. Because these unresolved motions can continue to influence subsequent dynamics, the resulting dynamics retain memory that is absent from a Markovian description. Here, we present ugme, a modular, open-source Python package for constructing and validating generalized master equations (GME) models that explicitly account for these non-Markovian (memory) effects.
ER -