QM_descriptors_calculation

This repository contains a QM descriptor generator which automatically calculate atomic/bond descriptors, e.g. partial charges, fukui indices, and bond index.

Requirements

While it is possible to run all of the code on a normal desktop, HPC makes the calculation significantly faster. To run the code, you will need:

1. python 3.7
2. rdkit 2019
3. pandas
4. GFN2-xtb by Grimme https://xtb-docs.readthedocs.io/en/latest/contents.html
5. NBO for population analysis https://nbo6.chem.wisc.edu/, for NBO configurations see: https://nbo6.chem.wisc.edu/INSTALL.gaussian
6. G16 for SCF calculation.

The first three can be installed through the following conda environment.

Installation

1. install Miniconda from https://conda.io/miniconda.html
2. create conda environment for QM_descriptors by:

   conda env create -f environment.yml

   source activate QM_descriptors

Usage

To calculate the QM descriptors, run: python main.py --smiles <input.csv>

For a list of optional flag, run: python main.py -h

Two paths need to be specified in the main.py:

XTB_PATH = '$GFN_XTB_PATH'
G16_PATH = '$G16_PATH'

Input

The code takes a .csv of smiles as input, e.g.:

id,smiles
0,CHEMBL231079,C
1,CHEMBL1521196,C1=CC=CC=C1

The .csv must have these two columns with the same head. The id can be either str or int.

Output

The code will generate three folders, as specified by the arguments '--MMFF_conf_folder', '--xtb_folder', '--DFT_folder', which hold results for MMFF conformer searching, semi-empirical optimization, and DFT electronic structure calculations, respectively.

The QM descriptors from DFT calculations will be parsed automatically and saved as a dataframe in the .pickle file specified by '--output' argument.

Use on the HPC

See submit.sh for an example of submitting scripts. The following parameters in the submit.sh need to be replaced:

$CONDA_PYTHON_PATH: python path in your conda environment, e.g.: /home/yanfeig/miniconda3/envs/QM_descriptors/bin
$NBOPATH: path to the NBO bin folder, e.g.: /home/yanfeig:/home/yanfeig/nbo6/bin
$CONDA_PACKAGE_PATH: package path in your conda environment, e.g.: /home/yanfeig/miniconda3/envs/QM_descriptors/lib/python3.7/site-packages
$G16ROOT: path to the G16 root
$SCRATCH_FOLDER_G16: Scratch folder holding G16 calculation 

Known limitations

Allowed atom types currently are limited to the MMFF94 force field, which is used to search conformers for the input molecules. Allowed atoms and ions are: C, H, N, O, F, Si, P, S, Cl, Br, I, Fe+2, Fe+3, F-, Cl-, Br-, Li+, Na+, K+, Zn+2, Ca+2, Cu+1, Cu+2, and Mg+2

Contributors

Yanfei Guan, Duminda Ranasinghe, Oscar Wu