Interpretable Machine Learning Disentangles Pore Confinement Across Multiple Zr-MOF CO2 Capture Targets
Abstract
Pore confinement in zirconium-based metal–organic frameworks (Zr-MOFs) is often described using the cage–window ratio (ΓCW), the largest cavity diameter divided by the pore-limiting diameter (Df), yet whether this composite coordinate carries information beyond Df alone remains unclear. We modeled CO2 uptake, heat of adsorption (HoA), and CO2/N2 selectivity for 239 Zr-containing MOFs at two pressures relevant to vacuum-swing adsorption using structure-grouped machine learning with SHAP interpretation, reaching test R2 values of 0.77–0.88. Although ΓCW correlated positively with all three targets, controlling for Df weakened these associations and reversed their sign, while the reciprocal analysis left the Df associations largely unchanged and a difference-based formulation produced no significant association. Df played different roles across the three targets, indicating that the apparent confinement effect mainly reflects the limiting aperture rather than relative cage–window architecture.
Related
- Machine Learning-Powered Simulations of Confinement-Driven ¹²⁹ Xe Nuclear Spin Relaxation
- Interpretable Machine Learning-Based QSPR Modeling for Predicting Hydrogen Adsorption Performance of Ester Oil Impregnating Agents
- Physics-inspired feature engineering and interpretable machine learning for thermoelectric properties of multicomponent materials
- Toward Interpretable Graph-Based Machine Learning for Combustion Kinetic Model Generation
- Enhancing Response Surface Methodology for Bioprocesses Development Using Interpretable Machine Learning
BibTeX
@misc{vali2026interpretable,
title = {{Interpretable Machine Learning Disentangles Pore Confinement Across Multiple Zr-MOF CO2 Capture Targets}},
author = {Leyli Vali and Seyed Majid Hashemianzadeh},
howpublished = {ChemRxiv},
year = {2026},
month = sep,
doi = {10.26434/chemrxiv.15008444/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008444/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Interpretable Machine Learning Disentangles Pore Confinement Across Multiple Zr-MOF CO2 Capture Targets
AU - Leyli Vali
AU - Seyed Majid Hashemianzadeh
PY - 2026
DA - 2026/09/07
PB - ChemRxiv
DO - 10.26434/chemrxiv.15008444/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008444/v1
AB - Pore confinement in zirconium-based metal–organic frameworks (Zr-MOFs) is often described using the cage–window ratio (ΓCW), the largest cavity diameter divided by the pore-limiting diameter (Df), yet whether this composite coordinate carries information beyond Df alone remains unclear. We modeled CO2 uptake, heat of adsorption (HoA), and CO2/N2 selectivity for 239 Zr-containing MOFs at two pressures relevant to vacuum-swing adsorption using structure-grouped machine learning with SHAP interpretation, reaching test R2 values of 0.77–0.88. Although ΓCW correlated positively with all three targets, controlling for Df weakened these associations and reversed their sign, while the reciprocal analysis left the Df associations largely unchanged and a difference-based formulation produced no significant association. Df played different roles across the three targets, indicating that the apparent confinement effect mainly reflects the limiting aperture rather than relative cage–window architecture.
ER -