A webserver for identifying protein binding sites of protein-nucleic acid complexes. ATMGBs (Attention Maps and Graph convolutional networks for Binding site prediction) is a novel deep learning framework for sequence-based prediction of protein-nucleic acid binding sites. It integrates multiple feature types and leverages attention maps from protein language models to simulate spatial relationships between residues. It needs one input: a sequence in fasta format. More details can be found in the tutorial.