Given the significant improvement in accuracy for calculation of the partition
coefficient for neutral species (logP) it was only natural to pose the question whether similar results would
occur for the dissociative partition coefficient, logD.
Two test cases were chosen. The first case was a comparison [2] of experiment vs.
theory for the octanol-water partition coefficient at pH 7.4 for a set of 18 benzylphenylpiperidines
that have a range of substituents at the 5-benzyl position; this is compound (V). The second case [3]
was a set of 10 compounds which were 2,3,4-substituted analogs of compound (VI).
The discrepancies of logD (partition coefficient in octanol-water) for the first
set of compounds were found to be larger than average, with an average discrepancy of 0.94 logD units.
The second set of compounds had a very high average discrepancy, of 1.29 logD units. A search through the
logP and pKa databases from which the logP and pKa prediction parameters
are derived prior to estimating logD
showed that both classes of compounds were under-represented.
The substituents which were added at the R position for compound (V) ranges
from simple halogens to sophisticated heterocycles, including thiazolyl- and pyrimidinyl- groups. The R1,
R2, and R3
substituents for compound (VI) were simply hydro-, methyl-, fluoro-, and chloro- groups.
The starting basis for the system training were the compounds (VII) and (VIII).
These compounds represented the simplest versions of (V) and (VI) for which experimental values were known:
The first operation consists of working backwards from the logD values to logP values.
As with the logP user training algorithm, once a compound is designated as a "training structure" for
logD user-training prediction, the algorithm breaks its structure down into the most important components,
and then re-scales its parameter estimation based on the newly-identified fragments. Both the production of
fragments and the re-scaling process are automatic and reproducible, and can be applied to any other case.
For the 5-substituted analogs of compound (V), the results for the logD calculation using
"generic" parameters (i.e., no specialization in the type of structures of interest) are summarized in Table 4.
The calculated logD value has an average discrepancy from experiment of 0.94 logP units.
The effect of system training with user-supplied data is also summarized in that table.
The average discrepancy has been reduced to 0.38 logD units.
|
