Hi RDKitters,
I am using the RDKit implementation of TFD to examine conformational
differences between macrocycles and to cluster their conformations. In some
basic testing of a set of conformations of the same macrocycle from the PDB I
find something unexpected:
#mollist = list of conformations of the same molecule from the PDB
tfds = []
rmsds = []
for m in mollist:
for n in mollist:
tfd = TorsionFingerprints.GetTFDBetweenMolecules(m, n, maxDev='spec',
useWeights=False)
rmsd = GetBestRMS(m,n)
tfds.append(tfd)
rmsds.append(rmsd)
#Plot the two lists
[image.png]
Comfortingly, all cases of RMS = 0 also give TFD = 0.
According to the original paper a TFD of 1 implies maximal torsional deviation,
yet here I see a very low RMSD (0.3-4A, essentially insignificant for molecules
of this size) at TFD = 1.
Also useWeights = True in the code above gives TFDs > 1, which is clearly not
possible in the spirit of the original idea, but probably arises from there not
really being a graph centre in a macrocycle.
The idea of clustering macrocycles based on some measure of distance in torsion
space is very appealing, but I am concerned by TFD = 1 being calculated for
conformations that have essentially the same geometry. Any suggestions on how
to proceed?
Paul.
Sent with [ProtonMail](https://protonmail.com/) Secure Email.
_______________________________________________
Rdkit-discuss mailing list
Rdkit-discuss@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/rdkit-discuss
