EMMI dataset

The EMMI (Enzyme–sMall Molecule Interaction) dataset contains experimentally confirmed enzyme–small molecule pairs. Data were integrated from multiple sources, including BindingDB, BRENDA, ChEMBL, Gene Ontology, IUPHAR-PSB, PubChem, Rhea, SABIO-RK, and UniProt. EMMI comprises 147,224 samples spanning three relation types: enzyme–non-interacting, enzyme–substrate, and enzyme–inhibitor pairs. It includes 15,652 unique enzymes and 28,977 unique small molecules, yielding an overall enzyme-to-small molecule ratio of 0.54. The dataset is well-balanced across interaction types while preserving diversity among both enzymes and small molecules.To better reflect biological reality—where most enzymes do not interact with most small molecules—the number of enzyme–non-interacting pairs (79,668) was kept higher than the combined number of enzyme–substrate and enzyme–inhibitor pairs (67,556). The ratio of enzyme–substrate to enzyme–inhibitor samples is approximately 1:1. The EMMI dataset is part of the broader EMMA project—a general transformer-based multi-task learning framework designed to predict different interaction types between enzymes and small molecules.