Re: [Open Babel] evaluate rmsd

[Craig A. J. <cj...@em...>]

nicolas-gaudin wrote:
> Hello,
> 
>  > On Jul 21, 2008, at 11:34 AM, Noel O'Boyle wrote:
>  >
>  > > My point is that if the molecule has some local symmetry, there might
>  > > be a better rmsd where you don't map the corresponding atoms. For
>  > > example, if a phenyl group in one conformer is rotated 180 with
>  > > respect to the phenyl in another conformer, the calculated RMSD should
>  > > be 0, but won't be if you just look at the corresponding atoms. Isn't
>  > > this right?
>  >
>  > Aah, yes. Got it
>  >
> 
> Yes, the problem is the atom mapping wich "destroy" all symetries, and 
> the naïve algorithm evaluate rmsd without change the mapping, resulting 
> in a huge estimation in the case of symetry. That's why we need to find 
> all automorphisms.


The trick is to use atom clases rather than simple one-to-one atom maps.  For example, a reactant that included NC(C)C would have just three classes, not four; both methyl atoms would have the same mapping number, and either methyl could map to the corresponding atom class on the right side of the reaction.

Using atom-class mapping, rather than atom-to-atom mapping, you don't artificially break the symmetry of the molecule when creating the atom map.

Craig