An Automated Method for Extracting Adsorption Binding Sites in MOFs from GCMC Simulations with Direct Comparison to Experimentally Determined Binding Sites

Abstract

The location and nature of adsorption binding sites in porous materials is central to understanding selective adsorption of guest molecules at the atomic level. While grand canonical Monte Carlo (GCMC) simulations routinely generate three-dimensional adsorbate probability densities, these data are most often interpreted qualitatively. The absence of a standardized and automated framework for extracting binding sites from adsorption probability distributions (APDs) limits quantification and analysis of binding sites in high-throughput screening workflows.

This thesis contains two related parts. The first is the development of a standalone, robust, and semi-automated workflow, termed GALP, for extracting adsorption binding sites from adsorbate probability densities generated by GCMC. GALP applies smoothing, peak identification, clustering, and molecular fitting to transform three-dimensional probability densities into discrete binding-site coordinates, along with binding-site-specific information such as relative occupancy and binding energetics. The development and validation of GALP establishes a general and transferable framework for systematic binding-site identification.

The second part of the thesis quantitatively evaluates the reliability of common classical simulation approximations used in GCMC, specifically the use of generic force field parameters within the framework, standard charge assignment schemes, and the rigid framework approximation. Simulation-predicted binding sites obtained under these assumptions are directly compared with experimentally determined adsorption sites. This comparison assesses the positional accuracy with which classical simulations reproduce known adsorption motifs, thereby identifying the conditions under which these approximations support meaningful mechanistic interpretation and the cases in which limitations become evident.

Together, these results demonstrate that probability-based binding site extraction provides a consistent and reproducible framework for validating classical simulation methodologies against experiment. By enabling direct, spatially resolved comparison of predicted and experimental adsorption sites, the approach moves beyond global adsorption metrics such as isotherms and uptake values, and instead focuses on the underlying structural features that govern adsorption behaviour. As a result, the methods developed here provide a physically grounded basis for scalable high-throughput analysis of adsorption in porous materials.