Machine learning of isomerization in porous molecular frameworks: exploring functional group pair distance distributions

Abstract

Molecular Framework Materials (MFMs), including Metal Organic Frameworks (MOFs), Covalent Organic Frameworks (COFs) and their discrete equivalents, Metal Organic Polyhedra (MOPs) and Porous Organic Cages (POCs) are porous materials, composed of molecular fragments, bound in one of many topologies. MFMs have a wide variety of potential and realised adsorption applications. In order to design an ideal framework material for a particular application, the composition of molecular fragments is not the only factor, but the arrangement of the those fragments is also important, especially when the fragments (molecular building blocks) are chemically functionalized and lack symmetry. As has been observed in metal organic frameworks, the flexibility and absorption properties may differ greatly when altering the orientation of the building units or changing the position of functional groups. However, although the position of the functional groups has a great influence on a targeted property, studies on functional group arrangements have only been performed on a small set of MOF structures. In this contribution, we develop a fingerprint/descriptor for optimising functionalized molecular framework structures using machine learning. We begin from the perspective of a molecular framework structure described as a collection of discrete pore shapes. To describe the chemical environment of the pore, we derive a fingerprint based on the occurrence of pairwise distances between functional groups in each pore. We present the possibilities of functional group arrangements in the 14 most common pore shapes, created by ditopic (2-connected) linkers. The method to enumerate and identify possible isomers is explained. Finally the performance of the fingerprint on predicting guest molecule binding energy is demonstrated.