ProLMS-GNN (Prediction of Ligand Molecule-binding Sites using Graph Neural Network) is a web server for predicting small molecule binding using a graph neural network model. Starting from a given PDB ID, UniProt ID, or structure data, ProLMS-GNN predicts ligand-binding residues and the binding ability for the ligand using simple structural features or embeddings computed by Evolutionary Scale Modeling (ESM). Currently, pyridoxal 5'-phosphate-like molecules (PLP), flavin-adenine dinucleotide-like molecules (FAD), nicotinamide adenine dinucleotide-like molecules (NAD), and Coenzyme A-like molecules (COA) binding prediction models have been prepared in addition to a general ligand binding residues prediction model.