Development and Bayesian Uncertainty Quantification of Transferable Coarse-Grained Models for Carbohydrates
- 1Department of Chemical Engineering, Virginia Tech, Blacksburg, Virginia 24061, United States
Abstract
Carbohydrates play an important role in industrial, biomedical, and materials applications. The exceptional structural diversity of carbohydrates makes their accurate molecular modeling both essential and challenging. Although coarse-grained (CG) molecular dynamics (MD) simulations provide access to large length and time scales relevant to carbohydrate assembly and materials behavior, the development of uncertainty-quantified transferable CG models remains challenging and limited. In this work, we employ a computational framework that integrates multiscale modeling, optimization algorithms, and machine learning to develop CG models that reproduce the structural, physical, and thermodynamic properties of 12 pyranose monosaccharides. To preserve their stereochemical and geometric features while balancing computational efficiency, each monosaccharide is represented using a six-bead mapping scheme, with each CG bead mapped based on the monosaccharide functional groups. To ensure transferability, force field (FF) parameters for CG models of three model monosaccharides, α-d-glucose, β-d-xylose, and α-l-rhamnose, were optimized using Particle Swarm Optimization (PSO). Bayesian uncertainty quantification was subsequently performed to assess the robustness of the optimized models by quantifying uncertainty in both the model parameters and the predicted properties. To assess parameter transferability, we applied the optimized CG beads developed for the model monosaccharides to other monosaccharides in CG MD simulations. The resulting CG models accurately reproduced thermodynamic and structural properties across 12 monosaccharides, demonstrating good transferability. Finally, the predictive capability of the optimized CG parameters was further assessed across 12 monosaccharides using glass transition temperatures and surface energies, which were not included during parameter optimization. The model reproduced glass transition temperature with good accuracy and captured qualitative surface-energy trends, demonstrating predictive capability beyond the optimized properties. Overall, this work establishes an uncertainty-aware CG modeling framework that enhances confidence in CG simulations and provides a scalable foundation for extending to oligosaccharides, polysaccharides, and hydrated carbohydrate assemblies.
Related
- Bayesian neural networks versus deep ensembles for uncertainty quantification in machine learning interatomic potentials
- NephroTox AI: an explainable machine learning framework with uncertainty quantification for drug-induced nephrotoxicity prediction
- PCAM: proteochemometric assay-aware modeling for robust bioactivity estimation and uncertainty quantification
- Development of a Deep Reinforcement Learning Approach for Transition State Search in Molecular Systems: A Transferable Model Across Structural and Dimensional Variations
- Towards Out-of-Distribution Uncertainty Quantification for Machine-Learned Interatomic Potentials via Pre-Training
BibTeX
@article{farzeen2026development,
title = {{Development and Bayesian Uncertainty Quantification of Transferable Coarse-Grained Models for Carbohydrates}},
author = {Parisa Farzeen and Soumil Y. Joshi and Abhishek T. Sose and Sanket A. Deshmukh},
journal = {Journal of Chemical Theory and Computation},
year = {2026},
month = oct,
doi = {10.1021/acs.jctc.6c01202},
url = {https://doi.org/10.1021/acs.jctc.6c01202},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Development and Bayesian Uncertainty Quantification of Transferable Coarse-Grained Models for Carbohydrates
AU - Parisa Farzeen
AU - Soumil Y. Joshi
AU - Abhishek T. Sose
AU - Sanket A. Deshmukh
PY - 2026
DA - 2026/10/08
JO - Journal of Chemical Theory and Computation
DO - 10.1021/acs.jctc.6c01202
UR - https://doi.org/10.1021/acs.jctc.6c01202
AB - Carbohydrates play an important role in industrial, biomedical, and materials applications. The exceptional structural diversity of carbohydrates makes their accurate molecular modeling both essential and challenging. Although coarse-grained (CG) molecular dynamics (MD) simulations provide access to large length and time scales relevant to carbohydrate assembly and materials behavior, the development of uncertainty-quantified transferable CG models remains challenging and limited. In this work, we employ a computational framework that integrates multiscale modeling, optimization algorithms, and machine learning to develop CG models that reproduce the structural, physical, and thermodynamic properties of 12 pyranose monosaccharides. To preserve their stereochemical and geometric features while balancing computational efficiency, each monosaccharide is represented using a six-bead mapping scheme, with each CG bead mapped based on the monosaccharide functional groups. To ensure transferability, force field (FF) parameters for CG models of three model monosaccharides, α-d-glucose, β-d-xylose, and α-l-rhamnose, were optimized using Particle Swarm Optimization (PSO). Bayesian uncertainty quantification was subsequently performed to assess the robustness of the optimized models by quantifying uncertainty in both the model parameters and the predicted properties. To assess parameter transferability, we applied the optimized CG beads developed for the model monosaccharides to other monosaccharides in CG MD simulations. The resulting CG models accurately reproduced thermodynamic and structural properties across 12 monosaccharides, demonstrating good transferability. Finally, the predictive capability of the optimized CG parameters was further assessed across 12 monosaccharides using glass transition temperatures and surface energies, which were not included during parameter optimization. The model reproduced glass transition temperature with good accuracy and captured qualitative surface-energy trends, demonstrating predictive capability beyond the optimized properties. Overall, this work establishes an uncertainty-aware CG modeling framework that enhances confidence in CG simulations and provides a scalable foundation for extending to oligosaccharides, polysaccharides, and hydrated carbohydrate assemblies.
SN - 1549-9618
SN - 1549-9626
ER -