Expanding Automated Multiconformer Ligand Modeling to Macrocycles and Fragments

Upon binding to a protein, ligand conformational flexibility is generally reduced. However, ligands can retain some conformational flexibility even in the bound state. In the Protein Data Bank (PDB), a small number of ligands have been modeled with distinct alternative conformations supported by X-ray crystallography density maps. However, the vast majority of structural models are fit to a single ligand conformation, potentially ignoring the underlying conformational heterogeneity present in the sample. To help model conformational heterogeneity, we have developed a new version of qFit, which can automatically capture conformational heterogeneity as supported by the underlying experimental data. The underlying concept of qFit is to enumerate a large number of conformations according to a sampling procedure and then to use mixed integer quadratic programming to optimize the selection of a parsimonious set of conformers, along with their corresponding occupancies. Here, we introduce how different sampling algorithms can be substituted into the qFit algorithmic infrastructure to identify multiple conformations of ligands, as supported by the underlying experimental data. Our new sampling strategies, driven by the Experimental-Torsion Knowledge Distance Geometry (ETKDG) conformer generator in RDKit, allowed us to simultaneously identify alternative conformations of ligands that improve the fit of experimental data and reduce torsional strain. We demonstrate identifying multiple conformations of complex ligands such as macrocycles and small fragments in PanDDA-modified density maps from high throughput X-ray. These advances enhance the analysis of residual conformational heterogeneity present in ligand-bound structures, with future diffusion-based sampling strategies increasing the identification of ligand heterogeneity in the PDB, providing important insights for the rational design of therapeutic agents.