UniSite: The First Cross-Structure Dataset and Learning Framework for End-to-End Ligand Binding Site Detection

UniSite represents a breakthrough in ligand binding site detection, a crucial aspect of Structure-Based Drug Design. Traditional methods have faced significant challenges, including a focus on individual protein-ligand complexes and inadequate evaluation metrics that fail to capture the true performance of prediction methods. The introduction of UniSite-DS, which contains 4.81 times more multi-site data and 2.08 times more overall data than previously used datasets, directly addresses these issues. By employing an end-to-end detection framework supervised by set prediction loss, UniSite aims to provide a more comprehensive approach to identifying binding sites across multiple complexes of the same protein. This advancement not only enhances the reliability of predictions but also supports the ongoing evolution of drug design methodologies, ultimately contributing to more effective therapeutic solutions.