ChEMBL3D: Quantum-Accurate 3D Conformers for ChEMBL at Scale

ChEMBL3D is a large-scale 3D molecular conformer dataset curated from ChEMBL v34. It provides hydrogen-complete molecular topologies and optimized conformer geometries, organized by atom-count groups for efficient access and processing.The dataset includes:topologies: grouped SDF/SMI files containing molecular topologies (group IDs such as 007, 008, …).zarr_database: size-grouped Zarr conformer data with aligned arrays for:atomic numbers (numbers)3D coordinates (coord)absolute energies (energy)relative energies (relative_energy)stereochemistry identifiers (stereo_id)molecule identifiers (mol_id)molecular charge (charge)Conformers were produced and curated through a multi-stage workflow including protomer/stereoisomer handling, conformer generation, geometry optimization with an AIMNet2 potential, and topology/conformer consistency checks. The release is intended to support molecular ML, conformer ranking, force-field development, and large-scale cheminformatics workflows.For distribution on platforms with file-count limits, this release is provided as compressed archives:topologies.tar.zstzarr_database.tar.zst.part-* (split compressed archive parts)Archive integrity checks are provided via ARCHIVE_SHA256SUMS.For full dataset structure, packaging/unpacking commands, and usage examples, see the included README.md.