Name: MSR-ACC-TAE25_Train Extended ID: MSR-ACC-TAE25_Train__Ehlert-Hermann-Vogels-Satorras-Lanius-Segler-Giesbertz-Kooi-Takeda-Huang-Luise-Berg-Gori-Giorgi-Karton__DS_yf2caw9gh4p7_0 Description: MSR-ACC/TAE25 (Microsoft Research Accurate Chemistry Collection, Total Atomization Energies 2025) provides 73,040 total atomization energies (TAEs) at the CCSD(T)/CBS level obtained with the W1-F12 composite wavefunction protocol implemented in Molpro 2024.1. This is the canonical training split comprising 71,871 molecules (99% of molecules remaining after removing overlap with the W4-17 and GMTKN55 benchmark sets).The dataset covers the chemical space of closed-shell, charge-neutral, covalently bound equilibrium molecular structures containing up to 5 non-hydrogen atoms drawn from elements H through Ar, excluding rare gases. Molecular structures were generated by exhaustive graph enumeration and degree-sequence sampling, then optimized through a cascade of GFN2-xTB, r2SCAN-3c, and B3LYP-D3(BJ)/def2-TZVPP levels of theory (ORCA). Structures were filtered to exclude those with significant multireference character (%TAE[(T)] > 6% at CCSD(T)/6-31G*), triplet electronic ground states, or dissociated fragments. The W1-F12 protocol includes Hartree-Fock extrapolation to the complete basis set limit (cc-pVDZ-F12 and cc-pVTZ-F12, alpha=5), CCSD-F12b correlation, perturbative triples delta(T) using jul-cc-pV(D+d)Z and jul-cc-pV(T+d)Z basis sets (alpha=3.22), and a core-valence correction using cc-pwCVTZ. The dataset spans 45.1% organic and 54.9% inorganic molecules and provides broader chemical diversity than comparable datasets such as GDB-9 or VQM24/DMC. Additional data available in the source files, including DFT atomization energies at approximately 90 levels of theory, singlet-triplet gaps, %TAE[(T)] multireference diagnostics, and W1-F12 energy components, can be downloaded from ColabFit Exchange. Authors: Sebastian Ehlert Jan Hermann Thijs Vogels Victor Garcia Satorras Stephanie Lanius Marwin Segler Klaas J.H. Giesbertz Derk P. Kooi Kenji Takeda Chin-Wei Huang Giulia Luise Rianne van den Berg Paola Gori-Giorgi Amir Karton DOI: None Calculated Property Types: atomization_energy Elements: Al (4.52%) B (5.61%) Be (3.08%) C (8.18%) Cl (1.03%) F (1.3%) H (44.58%) Li (2.18%) Mg (2.34%) N (6.34%) Na (1.41%) O (3.55%) P (5.34%) S (4.29%) Si (6.24%) Methods: W1-F12/CCSD(T)-CBS Software: Molpro 2024.1 Number of Configurations: 71,871 Number of Atoms: 532,242 Publication Link: https://doi.org/10.1038/s41597-026-07200-8 Data Source Link: https://huggingface.co/datasets/microsoft/msr-acc-tae25 Other Links: https://arxiv.org/abs/2506.14492 https://zenodo.org/records/15783826