The accurate prediction of host–guest complex geometries is critical for structure-based affinity prediction and the rational design of macrocyclic carriers including cyclodextrins (CD) and cucurbiturils (CB). Here, we benchmark eight docking scoring functions to as-sess their reproduction of crystallographic structures and generation thermally accessible poses suitable for ensemble-based analyses. Our results suggest that while docking functions—broadly intended for protein-ligand complexes—are generally limited by either their scoring function’s ability to capture orientation-sensitive interac-tions, or the effectiveness of their search algorithms to sample con-formational space. These problems are compounded by host-specific challenges, especially related to the role of electrostatic forces involved in CB complexation and CD’s feature multiple equivalent hydrogen-bond sites. From these results, we propose a practical workflow for structure prediction, using an approximate scoring function for initial pose generation followed by higher level calculations for the pose refinement.
Ettema et al. (Mon,) studied this question.