NNMol-IR : A Machine learning program for computing Infrared spectrum form 3D molecular structure.

Requirement

• tensorflow,
• tensorflow_probability,
• tensorflow_addons,
• nvidia-tensorrt (if you want to use GPU)

After installation of conda, and activation of your environnement, Type :

pip install tensorflow==2.12.0
pip install tensorflow_probability==0.19.0
pip install tensorflow_addons==0.20.0
pip3 install nvidia-tensorrt

Installation

Using git, under a terminal, type :

git clone https://github.com/allouchear/NNMol-IR.git

You can also download the .zip file of NNMP-IR : Click on Code and Download ZIP

Main programs

train.py

Using the database in Data (h5 format) directory, make the training
see train.inp and ./xtrain bash script. Please note that in this example, the IR data used are from a scaled DFT, not from experiment. In the published paper, we used the experimental data. However, we cannot share these values because they are from a commercial database. You can purchase them from NIST. See our published paper for more details. The data is not included in the GitHub repository. You have to download it from Zenodo: doi.org/10.5281/zenodo.13681778

evaluation.py

Test the models (one or an ensemble of models) using a database
see ./xevaluation bash script in example directory. Please note that the models are not included in the GitHub repository. You have to download them from Zenodo: doi.org/10.5281/zenodo.13681778

predict.py

Predict the IR spectrum using a xyz file
see ./xpredict* bash scripts. Please note that the models are not included in the GitHub repository. You have to download them from Zenodo: doi.org/10.5281/zenodo.13681778

Citations

Please cite : Neural Network Approach for Predicting Infrared Spectra from 3D Molecular Structure, Chemical Physics Letters 856 (2024) 141603. https://doi.org/10.1016/j.cplett.2024.141603

title = {Neural network approach for predicting infrared spectra from 3D molecular structure},
journal = {Chemical Physics Letters},
volume = {856},
pages = {141603},
year = {2024},
issn = {0009-2614},
doi = {https://doi.org/10.1016/j.cplett.2024.141603},
url = {https://www.sciencedirect.com/science/article/pii/S0009261424005451},
author = {Saleh {Abdul Al} and Abdul-Rahman Allouche},
keywords = {Infrared Spectra, Machine Learning, Neural Network, DFT scaled frequencies},
abstract = {We developed a machine learning (ML) model to directly predict IR spectra from three-dimensional (3D) molecular structures. The spectra predicted by our model significantly outperform those from density functional theory (DFT) calculations, even after scaling. In a test set of 200 molecules, our model achieves a Spectral Information Similarity Metric (SIS) of 0.92 surpassing the value achieved by DFT scaled frequencies which is 0.57. Additionally, our model considers anharmonic effects offering a fast alternative to laborious anharmonic calculations. Moreover, our model can be used to predict various types of spectra (as UV or NMR) as a function of molecular structure.}
}

Contributors

The code is written by Abdul-Rahman Allouche.

License

This software is licensed under the GNU General Public License version 3 or any later version (GPL-3.0-or-later).