Comparison of the Accuracy of LogP and LogD
Calculations for 22 Drugs

III. Results and Conclusions

Table 1 compares the experimental and calculated logD and pKa values for 22 drugs. Experimental logD values are listed at different pH. The calculated values have been obtained by the ACD/LogD and ACD/pKa programs.⁷⁾ The following correlations between the experimental and calculated values have been obtained:

LogD_Exp = (0.032±0.066) + (0.926±0.027)LogD_Calc (1)

N=50, R=0.9798, S=0.30

Figure 1 shows the corresponding data scatterplots. For the same drug data set the best correlation between the experimental and calculated logP values proved to be the following:¹⁾

LogP_Exp = (0.04±0.18) + (0.914±0.053))LogP_Calc (2)

N=22, R=0.9676, S=0.40

In previous work¹⁾ we suspected that experimental data for verapamil, haloperidol and diltiazem are incorrect. If we exclude them from correlations then we obtain the following results:

LogD_Exp = (0.042±0.055) + (0.954±0.024))LogD_Calc (3)

N=44, R=0.9873, S=0.25

LogP_Exp = (-0.04±0.13) + (0.979±0.042))LogP_Calc (4)

N=19, R=0.9843, S=0.30

Comparison of correlations (2) to (3) and (4) to (5) reveal that the logD parameter is predicted even more accurately than logP. This is quite unexpected result, since calculation of logD parameter is much more complex than calculation of just logP parameter. LogD prediction involves calculations of all individual logP values for all the ionic forms of drug molecule as well as all of its pKa values. Therefore one would expect that the resulting error of logD prediction must be much larger than the error of logP prediction for just neutral molecule.

In our opinion the only explanation of the observed results is that the experimental logP values are less accurate than the corresponding logD values. It is generally assumed that the most reliable logP values have to be obtained under such extreme pH values which suppress drug ionization, i.e. either pH>10 or pH<2. Under such extreme conditions drug partitioning may be affected by some effects which are not observed under the mild conditions at pH 5.0-8.0 where most of logD values are obtained. Therefore the logD parameter is calculated more accurately than the logP parameter. All these results reveal that logD is more useful for correlating biological properties of drugs than logP.