modelforge curated dataset: tmQM-xtb

Curated tmQM-xtb Dataset: T=100K dataset restricted to Pd, Zn, Fe, Cu, 1000 configuration test set: version: v1.1_PdZnFeCu_ntc1000_T100K

This dataset contains 103 unique systems with 1000 total configurations.  This dataset is limited to systems that contain transition metals Pd, Zn, Fe,  or Cu, and also only contain elements C, H, P, S, O, N, F, Cl, or Br. 

Potentially problematic configurations (i.e., unstable or those with underlying structural changes) were removed. Briefly, bond inference was performed on the initial configuration using RDKit and a configuration was excluded if any of those bond distances changed by more than 0.15 angstroms compared to the initial, energy minimized state. 

This dataset was generated starting from the tmQM dataset (release 13Aug2024, https://github.com/uiocompcat/tmQM) containing 108541 unique molecules; each molecule was evaluated using gfn2-xtb, and then a short MD simulation performed to provide additional configurations of the molecules. 

The tblite package was used to evaluate the energetic of the system using the gfn2-xtb formalism.

MD simulations were performed using the Atomic Simulation Environment (ASE), using the Langevin integrator 

Simulations were performed at 100K with a 1 fs timestep and 0.01 1/fs friction damping factor.  

In all trajectories, the first configuration corresponds to the energy minimized configuration reported in the original tmQM dataset.

100 steps were taken between snapshots,  with 10 total snapshots per molecule

During MD sampling, gfn2-xtb accuracy was set to 2; all reported properties were calculated at accuracy level 1. 

Scripts used to perform the sampling can be found https://github.com/chrisiacovella/xtb_config_gen

Properties included: 

atomic_numbers  

positions  

"per_atom"

"nanometer"

forces

"per_atom"

 "kilojoule_per_mole / nanometer"

partial_charges 

"per_atom"

"elementary_charge"

energies  

"per_system"

"kilojoule_per_mole"

dipole_moment_per_system  

"per_system"

 "elementary_charge * nanometer"

total_charge  

 "per_system"

"elementary_charge"

spin_multiplicities 

"per_system"

"dimensionless"

stoichiometry 

"meta_data"