Predicting the Host-Guest Binding Gibbs Free Energy for Anion Guests

Host-guest chemistry is an emerging approach in the design of functional molecular systems, with growing potential in areas such as targeted drug delivery, sensing, and adaptive materials. However, the development of new hosts and finding new guests for existing hosts remains experimentally challenging due to complex synthesis and the need for selective binding. To address this, we present a computational model that predicts Gibbs free energies of binding using a two-component machine learning model. Built on a carefully curated dataset of 247 host-guest systems from the literature, the model combines the electrostatic component of the complexes with three-dimensional molecular fingerprints and achieves an R² of 0.77 and a mean average error of 5.3 kJ per mol on an external test set. We demonstrate its applicability using newly acquired experimental data for a mixHC[8] macrocycle, showing that the approach can distinguish between strong and weak binders; thus, enable efficient high-throughput screening for host-guest discovery.