CGenFF Dissimilarity Score

The algorithm is pretty simple. The philosophy is atom types are perhaps a better way of indicating similarity than using the existing fingerprinting mechanisms. This helps map out accurately different atom types that have some actual physical meaning. The

Similarity is weighted in accordance to the CHARMM potential energy function. Each term is an appropriate classifier into a molecule where we have our bonded terms: bonds, angles, dihedrals, impropers, urea-bradley and our non-bonded: Coloumbic interactions and the Lennard-Jones "6-12" potential that accounts for Pauli Repulsion and London Disperson forces between two atoms.

A geographical representation of the classifiers can be found below:

For similarity we only take into account the bonded terms except Urea-Bradley, each weight is assigned from reverse order with higher weights being associated with bonds and decreases incrementally as we approach classifiers that include more atom terms.

Classifier	Atom Terms	Weight
Bonds	2	4
Angles	3	3
Dihedrals	4	2
Improper Dihedrals	4	2

Import

from global_chem_extensions import GlobalChemExtensions
ff = GlobalChemExtensions().forcefields()

Compare Two Molecules

The higher the score the more dissimilar it is between each other.

Code

dissimilar_score = ff.compute_cgenff_dissimilar_score(
   'global-chem/example_data/forcefield_parameterization/perfluorobutanoic_acid.str',
   'global-chem/example_data/forcefield_parameterization/perfluorohexanoic_acid.str',
)

print ("CGenFF Dissimilar Score: %s" % dissimilar_score)

Output

CGenFF Dissimilar Score: 26