Classifying Host-Guest Topology with Ion Mobility-Mass Spectrometry and Machine Learning

Elucidating the topology of host-guest complexes is essential for the rational design of supramolecular assemblies. Building on the recent success of data-driven approaches, we evaluate the combination of ion mobility-mass spectrometry (IMS-MS), density functional theory (DFT) featurization, and machine learning to predict and classify the binding modes of 1:1 complexes formed between cucurbit[6]uril (CB6) and diamine guests. Training a regression model with DFT-derived molecular descriptors and experimentally determined collisional cross sections (CCS) enables predicting the CCS of host-guest complexes with a diverse set of diamine guests. The predicted values naturally separate in two distinct groups corresponding respectively to inclusion and exclusion complexes, thereby enabling topology classification. This approach demonstrates that DFT-featurization and IMS-MS data capture well host-guest topology and provide a framework for the data-driven design of supramolecular assemblies.