MPLID: Membrane Protein-Lipid Interface Dataset

A large-scale dataset of experimentally validated lipid contact residues derived from experimentally determined structures in the Protein Data Bank. 100% EXPERIMENTAL LABELS - NO COMPUTATIONAL DATABASE DEPENDENCIES Dataset Statistics (v2.0.0) Proteins: 4,704 Total residues: 8,055,325 Contact residues: 80,439 Contact rate: 1.00% Sequence clusters: 813 (30% identity) Lipid codes recognized: 117 Train/Validation/Test Splits Train: 2,578 proteins, 4,907,696 residues Val: 1,051 proteins, 1,403,838 residues Test: 1,075 proteins, 1,743,791 residues Key Features Labels derived 100% from experimentally resolved lipids in PDB structures 4.0 Angstrom all-atom heavy-atom distance cutoff 4,704 proteins across all membrane protein classes Cluster-aware splits prevent data leakage Fully reproducible from public PDB data GitHub: https://github.com/omagebright/MPLID