Compare shapes of molecules



In [1]:

    
from __future__ import print_function, division, unicode_literals

import oddt
from oddt.shape import usr, usr_similarity
print(oddt.__version__)









    



0.4.1-12-ga7d79bd

We'd like to compare the shape of heroin with other molecules.

ODDT supports three methods of molecular shape comparison: USR, USRCAT and Electroshape.
USR looks only at the shape of molecule.
USR-CAT considers the shape and type of atoms.
Electroshape accounts for the shape and charge of atoms.
All those methods have the same API.
We will use USR, because it's the simplest and the fastest.



In [2]:

    
heroin = oddt.toolkit.readstring('smi', 
    'CC(=O)Oc1ccc2c3c1O[C@@H]4[C@]35CC[NH+]([C@H](C2)[C@@H]5C=C[C@@H]4OC(=O)C)C')

smiles = ['CC(=O)Oc1ccc2c3c1O[C@@H]4[C@]35CC[NH+]([C@H](C2)[C@@H]5C=C[C@@H]4OC(=O)Cc6cccnc6)C',
          'CC(=O)O[C@@H]1C=C[C@@H]2[C@H]3Cc4ccc(c5c4[C@]2([C@H]1O5)CC[NH+]3C)OC', 
          'C[N+]1(CC[C@@]23c4c5ccc(c4O[C@H]2[C@@H](C=C[C@@H]3[C@@H]1C5)O)OC)C', 
          'C[NH2+][C@@H]1Cc2ccc(c3c2[C@]4([C@@H]1CC=C([C@H]4O3)OC)C=C)OC',
          'CCOC(=O)CNC(=O)O[C@H]1C=C[C@H]2[C@H]3Cc4ccc(c5c4[C@]2([C@H]1O5)CC[NH+]3C)OCOC',
          'CC(=O)OC1=CC[C@H]2[C@@H]3Cc4ccc(c5c4[C@@]2([C@@H]1O5)CC[NH+]3C)OC',
          'C[NH+]1CC[C@]23c4c5cc(c(c4O[C@H]2[C@H](C=C[C@H]3[C@H]1C5)O)O)c6cc7c8c(c6O)O[C@@H]9[C@]81CC[NH+]([C@H](C7)[C@@H]1C=C[C@@H]9O)C']

molecules = [oddt.toolkit.readstring('smi', smi) for smi in smiles]

To compute the shape using USR we need the molecule's 3D coordinates.



In [3]:

    
heroin.make3D()
heroin.removeh()
for mol in molecules:
    mol.make3D()
    mol.removeh()

Now we can use the usr function.



In [4]:

    
usr_heroin = usr(heroin)
usr_heroin









    Out[4]:





array([  3.12823939,   1.59761882,   0.55976415,   3.20355535,
         2.18059039,   1.01127279,   6.30361271,   6.8250246 ,
       -10.16452217,   5.39781094,   5.79236984,   0.93677378])

USR represents shape with 12 descriptors, which summarize the distribution of atomic distances in the molecule. For more details see Ballester & Richards (2007).
USR-CAT and Electroshape use more descriptors, 60 and 15 respectively.

Let's see how similar it is to a different molecule.



In [5]:

    
usr_similarity(usr_heroin, usr(molecules[0]))









    Out[5]:





0.20657099769142923

The similarity function returns a number in range (0, 1], where a higher number means that the molecules are more similar and 1 means that the molecules have identical shapes.
All methods (USR, USR-CAT and Electroshape) use the same similarity function.

We will find a molecule similar to oxamide.



In [6]:

    
similar_mols = []
for i, mol in enumerate(molecules):
    sim = usr_similarity(usr_heroin, usr(mol))
    similar_mols.append((i, sim))



In [7]:

    
similar_mols.sort(key=lambda similarity: similarity[1], reverse=True)
similar_mols









    Out[7]:





[(2, 0.48911179609705718),
 (3, 0.45645597603277854),
 (1, 0.39947994224342542),
 (5, 0.39212419145828936),
 (0, 0.20657099769142923),
 (6, 0.16292525187745338),
 (4, 0.15072869043524584)]



In [8]:

    
heroin









    Out[8]:

Heroin



In [9]:

    
idx_most = similar_mols[0][0] 
molecules[idx_most]









    Out[9]:

The most similar molecule



In [10]:

    
idx_least = similar_mols[-1][0] 
molecules[idx_least]









    Out[10]:

The least similar molecule

Similarity between these molecules:



In [11]:

    
usr_most = usr(molecules[idx_most])
usr_least = usr(molecules[idx_least])
usr_similarity(usr_most, usr_least)









    Out[11]:





0.13678393398437561