DeepGlycanSite is a deep learning model designed to predict carbohydrate-binding sites on protein structures with high accuracy. It integrates geometric and evolutionary features of proteins into a deep equivariant graph neural network (EGNN) with a transformer architecture, outperforming previous state-of-the-art methods. The model effectively predicts binding sites for diverse carbohydrates and has been validated through mutagenesis studies, revealing the guanosine-5′-diphosphate-sugar-recognition site of an important G-protein coupled receptor (GPCR). This work demonstrates the potential of DeepGlycanSite in understanding carbohydrate-protein interactions and developing new therapeutics. The model's robust performance across different carbohydrate-binding site classes, including monosaccharide, disaccharide, oligosaccharide, and nucleotide binding sites, highlights its generalized applicability. Additionally, DeepGlycanSite can predict specific binding sites for query carbohydrates, as demonstrated in the case of GDP-Fuc binding to P2Y14, a functionally important GPCR. The method's effectiveness is further validated through experimental validation, including calcium mobilization assays and molecular dynamics simulations, providing insights into the molecular mechanisms underlying carbohydrate-regulated protein functions.DeepGlycanSite is a deep learning model designed to predict carbohydrate-binding sites on protein structures with high accuracy. It integrates geometric and evolutionary features of proteins into a deep equivariant graph neural network (EGNN) with a transformer architecture, outperforming previous state-of-the-art methods. The model effectively predicts binding sites for diverse carbohydrates and has been validated through mutagenesis studies, revealing the guanosine-5′-diphosphate-sugar-recognition site of an important G-protein coupled receptor (GPCR). This work demonstrates the potential of DeepGlycanSite in understanding carbohydrate-protein interactions and developing new therapeutics. The model's robust performance across different carbohydrate-binding site classes, including monosaccharide, disaccharide, oligosaccharide, and nucleotide binding sites, highlights its generalized applicability. Additionally, DeepGlycanSite can predict specific binding sites for query carbohydrates, as demonstrated in the case of GDP-Fuc binding to P2Y14, a functionally important GPCR. The method's effectiveness is further validated through experimental validation, including calcium mobilization assays and molecular dynamics simulations, providing insights into the molecular mechanisms underlying carbohydrate-regulated protein functions.