FunFOLD5: Combining Localised Templates and Structural Homology for Accurate Predictions of Biologically Relevant Protein-Ligand Interactions.

FunFOLD5 predicts protein-ligand binding sites with 81.2% accuracy across diverse ligands by combining local and global template searches.

• Why it matters: Accurate prediction of ligand identity, binding-site location, and conformation remains challenging, especially for uncharacterized proteins, limiting functional annotation and drug discovery efforts.
• What they did: The method integrates Foldseek-based template retrieval, curated ligand data, ligand superposition, and spatial clustering, improving upon its predecessor by incorporating local pocket-level searches alongside global fold similarity.
• The result: FunFOLD5's local targeted protocol enhances ligand identification, residue-level recall, and binding-pose accuracy, demonstrating robust performance on CASP15 and CASP16 targets and enabling large-scale protein-ligand interaction analysis.