Overview of xBind, including methodology highlights and supporting references for key applications.
xBind is a freely accessible interactive web server for cross-molecular protein binding-site prediction. It predicts residue-level binding sites for protein--protein, protein--DNA and protein--RNA interactions by integrating protein language model embeddings from ESM-2 with sequence- and structure-derived features in symmetry-aware E(3)-equivariant graph neural networks. xBind supports both structure-based and sequence-only submissions: users may provide a monomeric protein structure in PDB/mmCIF format or a single-chain amino-acid sequence in FASTA format, with an optional on-the-fly AlphaFold structure prediction step for sequence-only inputs. The server returns residue-level binding probabilities with user-adjustable thresholds and provides interactive sequence and 3D structure visualizations, per-residue likelihood plots and downloadable result files for downstream analysis. xBind also includes job tracking, configurable privacy settings and embedded documentation to support reproducible and user-friendly usage by both computational and experimental researchers.
xBind delivers an end-to-end web platform that integrates protein language model embeddings, structure-aware feature fusion, and interactive visualization for cross-molecular binding-site prediction and interpretation.
Read articleEquiPNAS couples protein language models with equivariant GNNs to improve protein–DNA/RNA interface detection. xBind extends its feature fusion recipe to support both structure uploads and sequence-only jobs.
Read articleEquiPPIS leverages E(3) equivariant graph neural networks to deliver robust protein–protein interaction site prediction. Its geometric inductive biases inspired xBind’s privacy-aware queue and residue probability reporting.
Read article