Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/thomasjewson/molecular3dlengthdescriptors

A 3D conformational based molecular descriptor set for use in QSPR and Machine Learning.
https://github.com/thomasjewson/molecular3dlengthdescriptors

3d cheminformatics chemistry conformer crystallography descriptor learning machine molecule rdkit

Last synced: 4 months ago
JSON representation

A 3D conformational based molecular descriptor set for use in QSPR and Machine Learning.

Awesome Lists containing this project

README 

        
# Molecular3DLengthDescriptors



### Introduction



The **Molecular3DLengthDescriptors** package can do the following:



1. Calculates the lowest energy 3D conformer of a molecule using RDKit's UFF and MFF94 force fields.

2. Converts an array of coordinates into a set of lengths: longest, medium and shortest.

3. Descriptors can be calculated from these lengths.



### Applications



You can use this tool to do the following:



- Calculate these descriptors for a machine learning project from 2D SMILES input.

- Calculate these descriptors using crystallogaphic molecular coordinates exported from a database such as the CSD.

- Calculate these descriptors using coordinates calculated via another method (eg: GROMACS, BALLOON, CONFAB... etc)



### Installation



Installation of this package requires Python 3+. 



You can install this package two ways...



via pip:



`pip install Molecular3DLengthDescriptors`



(https://pypi.org/project/Molecular3DLengthDescriptors/)



Or manually, by placing the Molecular3DLengthDescriptors folder within 



`~/.local/lib/pythonX.X/site-packages`



The `rdkit` and `numpy` libraries are dependencies for this package. Install them with the following:



`conda install numpy`



https://www.rdkit.org/docs/Install.html



### How Molecular3DLengthDescriptors module works



![png](Molecular3DLengthDescriptors_pictures/HowMolecular3DLengthDescriptorsWorks.png)



1. Import package



```python

import Molecular3DLengthDescriptors as md

```

    



2. Calculate 3D coordinates of 2D molecule



This code is based off Wicker et al's nConf20 descriptor (10.1021/acs.jcim.6b00565).



```python

from rdkit.Chem import MolFromSmiles



smiles = "O=c1cccc(/C=C/c2ccccc2)o1"

mol_2D = MolFromSmiles(smiles)

mol_2D

```



![png](Molecular3DLengthDescriptors_pictures/Molecule.png)



```python

cc = md.Lengths.ComputeCoordsFrom2D()

coordinates = cc.GetCoords(mol_2D)

coordinates

```



    array([[ 4.19392625, -2.10217421,  1.20758067],

           [ 3.77614639, -1.12343758,  0.59832016],

           [ 4.72136167, -0.18159364, -0.02926615],

           [ 4.27935864,  0.88880918, -0.69605956],

           [ 2.87057618,  1.13819338, -0.81141094],

           [ 2.00047091,  0.2903415 , -0.24586024],

           [ 0.56197146,  0.46799511, -0.31472371],

           [-0.34101042, -0.35981258,  0.23919665],

           [-1.81020273, -0.21901879,  0.19470342],

           [-2.47547343,  0.83154283, -0.45235907],

           [-3.87295236,  0.9032948 , -0.45516855],

           [-4.62630912, -0.07414675,  0.18888447],

           [-3.98359642, -1.12449045,  0.8365183 ],

           [-2.58741845, -1.19525503,  0.83872962],

           [ 2.44345336, -0.85157198,  0.4663684 ]])



3. Calculate the lengths of the 3D molecule



The covariance of the molecular coordinates is calculated. Then eigenvectors and eigenvalues from the covariance matrix is calculated. The eigenvectors will be aligned in directions of most variance, least variance and right angles to both of those. Therefore, if the square root of the eigenvalues is taken these values will now be back on the same unit scale as the coordinates and hence can be interpreted as the lengths of the molecule. 



```python

lengths = md.Lengths.GetLengthsFromCoords(coordinates)

lengths

```



    [0.0001053619852081641, 1.124146300889793, 3.3578154223541476]



4. Calculate descriptors



You can calculate individual descriptors:



```python

print(md.Descriptors.Flatness(lengths))

print(md.Descriptors.Cubeularity(lengths))

print(md.Descriptors.Plateularity(lengths))

print(md.Descriptors.ShortOverLong(lengths))

print(md.Descriptors.MediumOverLong(lengths))

```



    0.0001053619852081641

    1.0504929488912333e-05

    35825.784590642164

    3.1378134875053776e-05

    0.334785019273531

    



Or all at once:



```python

md.Descriptors.CalcAllDesc(lengths)

```



    {'Flattness': 0.0001053619852081641,

     'Cubeularity': 1.0504929488912333e-05,

     'Plateularity': 35825.784590642164,

     'ShortOverLong': 3.1378134875053776e-05,

     'MediumOverLong': 0.334785019273531}



### Explainations of descriptors



To explain what the descriptors mean, place `help(...)` around the function to output the docstring.



```python

help(md.Descriptors.ShortOverLong)

```



    Help on function ShortOverLong in module Molecular3DLengthDescriptors.Descriptors:

    

    ShortOverLong(lengths)

        A measure of how cubic a molecules is.

        

        0 means either a needle or plate shape.

        1 means perfect cube

        

        ShortOverLong = Shortest / Longest

    

    



```python

help(md.Descriptors.CalcAllDesc)

```



    Help on function CalcAllDesc in module Molecular3DLengthDescriptors.Descriptors:

    

    CalcAllDesc(lengths)

        Flattness => Shortest length

            Measure of how flat a molecules is.

        

        Cubularity => (Shortest*Medium*Longest)/(Longest)**3

            1 is a perfect cube. 

        

        Plateularity => (Medium*Longest)/ Shortest

            Larger value, more 2D plate-like.

            

        ShortOverLong => Shortest / Longest

            1 is perfect cube, 0 is needle or plate-like

            

        MediumOverLong => Medium / Longest

            1 is perfect plate-like shape, 0 is perfect needle shape

    

    



### Reference



If you use this package please feel free to reference us with:



Jewson, T., & Cooper, R. I. (2021). Molecular3DLengthDescriptors (1.0.0). https://github.com/ThomasJewson/Molecular3DLengthDescriptors



Or add this BibTeX entry to your .bib library:

```

@misc{Jewson2021,

author = {Jewson, Thomas and Cooper, Richard I.},

title = {{Molecular3DLengthDescriptors}},

url = {https://github.com/ThomasJewson/Molecular3DLengthDescriptors},

year = {2021}

}

```