FoXSight: Facilitating Computational Hit Identification in Crystallographic Fragment Screening Experiments

X-ray crystallographic fragment screening reveals how small molecules bind to proteins, providing important starting points for drug design. Identifying these weak binding events remains difficult when electron density signals are noisy, partially occupied or obscured by solvent, and campaigns may comprise thousands of datasets. Here, we present FoXSight, a computational method for hit identification in crystallographic compound screening. FoXSight applies Difference-of-Gaussian (DoG) filtering to multi-reference subtraction maps to detect recurring, target-specific density features associated with bound ligands. Its output maps remain visually interpretable and support subsequent ligand modeling. Across six annotated campaigns comprising 2,715 datasets, the top 10% of each FoXSight ranking contained 134 of 264 confirmed hits. FoXSight served as the primary hit identification tool for a human Cathepsin L campaign with non-isomorphous crystals, where the output of established methods was impractical to inspect systematically. The FoXSight computations for its 1,481 datasets took 47 minutes on 75 CPU cores.