A computed thermoelectric feature database for 50,992 GNoME materials (v1)

A machine-learning-ready feature set of physical properties computed for 50,992 novel crystalline materials from Google DeepMind's GNoME database, oriented toward thermoelectric screening. For each material it provides composition and structure descriptors, CHGNet universal-potential energetics, magnetism, and a 64-dimensional learned crystal embedding, and — the distinctive part — a complete lattice-thermal-transport layer: elastic constants, Debye temperature, sound velocities, Cahill/Clarke minimum-kappa bounds, the Grueneisen parameter (via a CHGNet equation-of-state), and a Slack-model lattice thermal conductivity. To our knowledge this thermal-transport feature layer has not previously been computed and released for the GNoME set. Files: (1) gnome_thermoelectric_features.parquet — main table, one row per material, all scalar features plus the 64-d embedding; (2) gnome_elastic_tensors.parquet — full 6x6 Voigt elastic tensor per material; (3) gnome_relaxed_structures.parquet — CHGNet-relaxed structure (CIF) per material. See README.md and SCHEMA.md inside the archive for methods, validation, and the full column dictionary. Derived from GNoME (Google DeepMind), which is licensed CC BY-NC 4.0; this derived dataset is released under the same NonCommercial terms.