Assessment of GFN-FF and GFN2-xTB methods for binding free energy calculations: Application to cucurbit[7]uril-cationic guest complexes

Abstract

The performance of two cost-efficient computational methods, the semiempirical extended tight-binding GFN2-xTB and the GFN-FF force field, was assessed for predicting the binding free energies of cucurbit[7]uril host–guest complexes with a series of cationic ammonium guests. Binding energies were computed via the supermolecular approach and decomposed into gas-phase interaction, thermostatistical, and solvation components. Both methods reproduce the general energetic trends of experimental binding affinities, with GFN-FF yielding lower mean absolute errors and standard deviations. However, the GFN2-xTB geometries are consistently closer to high-level DFT reference structures. The results indicate that while GFN-FF offers substantial computational efficiency and acceptable accuracy for screening purposes, its apparent energetic agreement may arise partly from error cancellation rather than superior physical description. The study also highlights the limitations of the single-structure supermolecular approach, particularly for flexible host–guest systems, where conformational sampling are required for a quantitatively reliable description of binding thermodynamics.