I think it depends on what you need the descriptor for. If it were for some
kind of fingerprinting, the example implementation would be too noisy. We
used it to estimate how many low energy conformations of a molecule might
be present in a particular system - and it turned out that correlated well
with our classifications of the system. The variability increases with
RBC: for totally rigid systems RBC and nConf20 are zero. For more
reproducible results you can increase the number of conformers generated;
the cost is longer calculations, but if you only have 350 molecules this
might be OK.
In the paper there are two example molecules with RBC of 1 and 8
respectively which both have only a single low energy conformation, and it
was this discrimination beyond simple RBC that drove its development.
Analysis of the spread of nConf20 showed that it was larger than the spread
of RBC, which might give it slightly better properties as an input
descriptor. However, if you are finding less variability in your particular
data set, then it might not be such a good discriminator of whatever you're
trying to discriminate. I wouldn't recommend adopting it as the 'main
descriptor' until you test whether it's useful.
Regards,
Richard
On Wed, Aug 29, 2018 at 3:24 PM Ali Eftekhari <a.b.eftekh...@gmail.com>
wrote:
> Hi Dr. Cooper,
>
> Thanks for your response and the suggestions. I added randomSeed=737 and
> I now get value of 14 for descriptor nConf20 for ZINC000290539224 molecule
> (although it is different than your paper [the value is 10] it does not
> change on each run). My concern now is on the general usage of nConf20
> descriptor. For instance, is there a limitation on what molecules can be
> used for estimating their nConf20? Since the conformers are generated
> randomly, how reliable is this descriptor to use it as a replacement for
> Rotatable Bond Count (RBC) in all machine learning models.
>
> In my application, the calculated values of RBC for 350 molecules range
> from 0 to 7 with (80% between 0-4 and 20% between 5-7). The calculated
> values of nconf20 is between 0-40 but with 95% between 0-3. Since nConf20
> for majority of molecules is between 0-3, I am concerned on the usage of
> nconf20 as the main descriptor. Could you please comment on that?
>
> Thanks,
> Ali
>
> On Wed, Aug 29, 2018 at 6:32 AM Richard Cooper <
> richardiancooper+rdkitdisc...@gmail.com> wrote:
>
>>
>> Just to follow up with the details - here is the line in the script to
>> change:
>>
>> conformers = AllChem.EmbedMultipleConfs
>> (molecule,numConfs,pruneRmsThresh=0.5, numThreads =3)
>>
>> to
>>
>> conformers = AllChem.EmbedMultipleConfs
>> (molecule,numConfs,pruneRmsThresh=0.5, numThreads =3, randomSeed=737 )
>>
>> (where 737 is an integer constant of your choice, but not -1).
>>
>> Richard
>>
>>
>> On Tue, Aug 28, 2018 at 12:55 PM Richard Cooper <
>> richardiancooper+rdkitdisc...@gmail.com> wrote:
>> >
>> > Hi Ali,
>> >
>> > Sorry I missed your email.
>> >
>> > The behaviour you describe is correct, due to a random seed in the
>> conformer generation step. The descriptor value usually doesn't vary by too
>> much.
>> >
>> > I think you can give the conformer generation a constant random seed if
>> you need a reproducible number for nConf20.
>> >
>> > Regards, Richard
>> >
>> >
>> > On Tue, 28 Aug 2018, 00:25 Ali Eftekhari, <a.b.eftekh...@gmail.com>
>> wrote:
>> >>
>> >> Hello all,
>> >>
>> >> I am trying to calculate 3D Descriptors following this publication:
>> >> "Beyond Rotatable Bond Counts: Capturing 3D Conformational Flexibility
>> in a Single Descriptor", Jerome G. P. Wicker and Richard I. Cooper. J.
>> Chem. Inf. Model. 2016, 56, 2347−2352
>> >>
>> >> I am essentially using the same script as they have in the supporting
>> information and i have attached it here as well. In Table 2 from the above
>> calculation, the value of the descriptor (nConf20) for ZINC000290539224
>> molecule is listed as 10. However, when I run the exact code as the one
>> they used, I get different value at each run.
>> >>
>> >> I have already contacted the authors but got no response. I am
>> wondering if the code they have in the supporting information is not right
>> or the value they listed in the table is wrong?
>> >>
>> >> The SMILES string for this particular molecule is:
>> >> 'CC(C)N2CC(NCc1cnc(C(C)O)s1)CC2=O'
>> >>
>> >> Thanks in advance for your help!
>> >>
>>
>>>
------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
Rdkit-discuss mailing list
Rdkit-discuss@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/rdkit-discuss
