Learning Multiscale Computational Strategies for Supramolecular Recognition
12 affiliations
- 1School of Computer Science and Engineering
- 2Beihang University
- 3Faculty of Synthetic Biology
- 4Shenzhen University of Advanced Technology
- 5College of Life Sciences
- 6Nankai University
- 7Department of Biomedical Engineering
- 8Southern University of Science and Technology
- 9School of Physics and Technology
- 10University of Jinan
- 11School of Physical Science and Technology
- 12Yangzhou University
Abstract
Accurate prediction of host-guest binding affinities with end-point free energy approaches requires a balance between molecular sampling, energetic description, and computational cost, yet the relative importance and interplay of these factors remain poorly understood for chemically diverse supramolecular systems. Here, we establish a systematic benchmark of end-point affinity calculations across more than 300 host-guest pairs involving pillararenes, cucurbiturils, octa acid, and cyclodextrins. We systematically evaluate single-(1T) and three-trajectory (3T) sampling, harmonic entropy corrections, multiple extended tight-binding and DFT levels, and implicit-solvent treatments, and then compare them with established MM/GBSA protocols. We find that different electronic-structure treatments can produce substantially different, host-dependent affinity landscapes, underscoring that the nominal level of theory alone is insufficient to determine predictive reliability. In contrast, the 3T scheme with independent sampling of bound and unbound states consistently improves affinity prediction, whereas MM-based normal-mode entropy provides little systematic benefit. A multi-objective Pareto analysis further reveals a pronounced accuracy-cost trade-off: although selected DFT protocols achieve high predictive accuracy, MM/GBSA 2 / 26 provides a remarkably favorable and broadly transferable balance between accuracy and computational expense. Finally, host-wise analysis and interpretable decision-tree modeling reveal systematic dependence of protocol preference on readily accessible host descriptors, particularly host charge and size. These results establish a practical framework for moving from universal method benchmarking toward host-aware selection of end-point computational protocols, providing a basis for more efficient and adaptive affinity prediction in supramolecular chemistry.
Related
- Site-selective methylene C–H oxidation of an alkyl diamine enabled by supramolecular recognition using a bioinspired manganese catalyst
- Structural, supramolecular and computational insights into the ALS inhibitor pyroxsulam: crystal structure, reactivity and bioactivity
- Supramolecular soft matter in traditional Chinese medicine: multiscale structure–function evidence, critical limitations, and translational challenges
- Advances in Multiscale Computational Modeling of 2D Nanomaterials for Solar Cells: From Interface Design to Fabrication and Stability
- Degradation Behavior of NBR Seals Exposed to Halogenated Olefins under Coupled Thermal and Mechanical Loads: A Multiscale Experimental and Computational Study
BibTeX
@misc{tang2026learning,
title = {{Learning Multiscale Computational Strategies for Supramolecular Recognition}},
author = {Wenting Tang and Zhenhua Lu and Lizhen Huo and Zhixuan Huang and Yang Li and Zuo-yuan Zhang and Zhaoxi Sun},
howpublished = {ChemRxiv},
year = {2026},
month = sep,
doi = {10.26434/chemrxiv.15008308/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008308/v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Learning Multiscale Computational Strategies for Supramolecular Recognition
AU - Wenting Tang
AU - Zhenhua Lu
AU - Lizhen Huo
AU - Zhixuan Huang
AU - Yang Li
AU - Zuo-yuan Zhang
AU - Zhaoxi Sun
PY - 2026
DA - 2026/09/04
PB - ChemRxiv
DO - 10.26434/chemrxiv.15008308/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008308/v1
AB - Accurate prediction of host-guest binding affinities with end-point free energy approaches requires a balance between molecular sampling, energetic description, and computational cost, yet the relative importance and interplay of these factors remain poorly understood for chemically diverse supramolecular systems. Here, we establish a systematic benchmark of end-point affinity calculations across more than 300 host-guest pairs involving pillararenes, cucurbiturils, octa acid, and cyclodextrins. We systematically evaluate single-(1T) and three-trajectory (3T) sampling, harmonic entropy corrections, multiple extended tight-binding and DFT levels, and implicit-solvent treatments, and then compare them with established MM/GBSA protocols. We find that different electronic-structure treatments can produce substantially different, host-dependent affinity landscapes, underscoring that the nominal level of theory alone is insufficient to determine predictive reliability. In contrast, the 3T scheme with independent sampling of bound and unbound states consistently improves affinity prediction, whereas MM-based normal-mode entropy provides little systematic benefit. A multi-objective Pareto analysis further reveals a pronounced accuracy-cost trade-off: although selected DFT protocols achieve high predictive accuracy, MM/GBSA 2 / 26 provides a remarkably favorable and broadly transferable balance between accuracy and computational expense. Finally, host-wise analysis and interpretable decision-tree modeling reveal systematic dependence of protocol preference on readily accessible host descriptors, particularly host charge and size. These results establish a practical framework for moving from universal method benchmarking toward host-aware selection of end-point computational protocols, providing a basis for more efficient and adaptive affinity prediction in supramolecular chemistry.
ER -