Published January 17, 2025 | Version 1.2
Model Open

XPaiNN models trained on QM9, SPICE, and SPICE-v2

  • 1. ROR icon Fudan University

Description

XequiNet-0.3 is the version of the publication. The repository is available at https://github.com/X1X1010/XequiNet/tree/v0.3.6

QM9: Contains XPaiNN models and configuration files trained on the QM9 dataset, including Δ-learning models for properties such as U0, U, H, and G. It also includes an HDF5 dataset of QM9.

SPICE: Contains the general organic molecular force field model trained on SPICE dataset.

    • xpainn-spice.pt: Parameter file of the direct-learning model. (Å, eV)
    • xpainn-spice.jit: JIT scripted file of the direct-learning model. (Bohr, Hartree)
    • xpainn@xtb-spice.pt: Parameter file of the Δ-learning model. (Å, eV)
    • xpainn@xtb-spice.jit: JIT scripted file of the Δ-learning model. (Bohr, Hartree)

ROT34: Contains xyz structures in ROT34 dataset.

    • ROT34.xyz: Standard structures given by the original publication, https://doi.org/10.1002/jcc.23649.
    • rot34_wb97m-d3bj_def2-tzvppd.xyz: Structures optimized by ωB97M-D3(BJ)/def2-TZVPPD.
    • rot34_xpainn.xyz: Structures optimized by XPaiNN model trained on SPICE.
    • rot34_xpainn@xtb.xyz: Structures optimized by XPaiNN@xTB Δ-learning model trained on SPICE.

XequiNet-1.2 is the newer version, whose repository is available at https://github.com/X1X1010/XequiNet

SPICE: Contains the models from XequiNet-0.3 but compatible with newer code version. It also contains model trained on SPICE-v2 dataset with net charge embedding.

    • xpainn-spice-v1.pt: Parameter file of the direct-learning model training on SPICE-v1. (Å, eV)
    • xpainn-spice-v1.jit: JIT scripted file of the direct-learning model training on SPICE-v1. (Å, eV)
    • xpainn@xtb-spice-v1.pt: Parameter file of the Δ-learning model on SPICE-v1. (Å, eV)
    • xpainn@xtb-spice-v1.jit: JIT scripted file of the Δ-learning model on SPICE-v1. (Å, eV)
    • xpainn-spice-v2.pt: Parameter file of model training on SPICE-v2, including net charge. (Å, eV)
    • xpainn-spice-v2.pt: JIT scripted file of model training on SPICE-v2, including net charge. (Å, eV)
    • xpainn-spice-v2-dipole.pt: Parameter file of model training on dipole moment label from SPICE-v2, including net charge. (Å, Debye)
    • xpainn-spice-v2-dipole.jit: JIT scripted file of model training on dipole moment label from SPICE-v2, including net charege. (Å, e*Å)
    • xpainn-spice-v2-dipole.pt: Parameter file of model based on xpainn-spice-v2-dipole.pt with an extra head for fitting atomic wise MBIS charges from SPCIE-v2. (Å, eV, e)

Please cite:

J. Chem. Theory Comput. 2024, 20, 21, 9500–9511

Files

XequiNet-0.3.zip

Files (2.3 GB)

Name Size Download all
md5:eb139f0a1170e3077c254303d46fa460
2.3 GB Preview Download
md5:c0e2712fedfb87baf8cbc37e06a5dd77
44.4 MB Preview Download

Additional details

Dates

Available
2025-01-17

Software

Repository URL
https://github.com/X1X1010/XequiNet
Programming language
Python