OpenFF Organometallic Complexes: Architector Minimum Energy Structures with Variable Ligand Subtypes

Description This dataset contains 22,343 mononuclear organometallic complexes generated using a space-filling categorical design over metal identity, oxidation state, coordination number, coordination geometry, and ligand composition. Structures were built and geometry-optimized with Architector using GFN-FF for initial assembly followed by GFN2-xTB optimization via TBLite and ASE. Each record corresponds to a single minimized geometry taken directly from the Architector output — no additional conformer sampling was performed. Ligands use the neutral (N), withdrawing (W), and donating (D) subtypes, in which dangling bonds are capped respectively with methyl groups, fluorine, or hydrogen. Candidates were generated from a stratified Farthest-Point Sampling (FPS) design spanning 15,000 target configurations for each subtype. Structures were retained if their coordination geometry and metal–ligand connectivity were maintained after optimization and if the total charge satisfied |q| ≤ 1. There are 5,603 entries with identical CMILES but with variations in oxidation state, total charge, and geometry. Coverage Property Values Metals Cu (2,615), Fe (5,161), Li (1,261), Mg (2,609), Pd (6,913), Zn (3,784) Charge -1 (12,856), +0 (5,679), +1 (3,808) Multiplicity 1 (7,280), 2 (5,538), 3 (1,852), 4 (1,232), 5 (4,140), 6 (2,301) Coordination Geometries axial_bicapped_hexagonal_planar, axial_bicapped_trigonal_prismatic, bent_109, bicapped_trigonal_prismatic, capped_octahedral, capped_square_antiprismatic, capped_trigonal_prismatic, dodecahedral, hexagonal_bipyramidal, hexagonal_planar, hexagonal_pyramidal, linear, octahedral, penta_bi_capped_pyramidal, pentagonal_bipyramidal, pentagonal_planar, pentagonal_pyramidal, seesaw, single, square_antiprismatic, square_planar, square_prismatic, square_pyramidal, t_shaped, tetrahedral, tri_tri_mer_capped, tricapped_trigonal_prismatic, trigonal_bipyramidal, trigonal_planar, trigonal_prismatic, trigonal_pyramidal Ligand Types Base Ligands: amine, ammonia, halide, hydrogen sulfide, hydroxyl, imido, methanediide, methyl, oxo, phosphido, phosphine, sulfido, thiol, waterLigand Subtypes: neutral (N: methyl capped), withdrawing (W, fluorine capped), donating (D, hydrogen capped) Oxidation states 0 (1,391), 1 (6,580), 2 (6,137), 3 (4,589), 4 (3,646) Coordination numbers 1 (439), 2 (1,572), 3 (2,721), 4 (2,575), 5 (2,470), 6 (2,926), 7 (2,733), 8 (3,817), 9 (1,847), 10 (618), 12 (625) Unique CMILES 18,859 Unique Stoichiometries 17,460 Total complexes 22,343 HDF5 Format The dataset is stored in modelforge HDF5 schema 2 format. Each complex is stored as an HDF5 group keyed by its `state_key` (a string encoding metal, oxidation state, coordination number, geometry, and ligand labels). Per-complex datasets follow the schema 2 conventions: `atomic_numbers` as `(n_atoms, 1)`, per-atom arrays as `(n_configs, n_atoms, 3)` (here n_configs=1), and per-system arrays as `(n_configs, 1)`. Each dataset carries `format`, `u` (openff-units string), and `property_type` attributes as required by the schema. Dataset Shape Units Format atomic_numbers (n_atoms, 1) — atomic_numbers positions (1, n_atoms, 3) nanometer per_atom xtb_energy (1,1) kilojoule_per_mole per_system total_charge (1,1) elementary_charge per_system spin_multiplicities (1,1) — per_system metal scalar string — meta_data geometry_type scalar string — meta_data oxidation_state scalar int32 — meta_data coordination_number scalar int32 — meta_data