Reliable Assessment of Log P of Compounds of Pharmaceutical Relevance

Abstract

In an earlier study Quant. Struct.-Act. Relat. 1996, 15, 403–409 comparing the performance of 14 log P predictors it was concluded that predictions of log P values were significantly better for simple organic molecules than for drugs. Since the publication of this benchmark study, a log P predictor, VLOGP, has been developed in our group. In the work presented here. VLOGP is used to assess the log P values of the same 48 drugs as included in the benchmark comparison. VLOGP returned 79.2% “acceptable”. 18.6% “disputable”, and only 2.2% “unacceptable” log P values. The “acceptable”, “disputable”, and “unacceptable” log P values from the 14 other predictors, respectively, ranged between 27.1% and 72.9%, 16.7% and 41.7%, and 2.2% and 37.5%. Further, VLOGP resulted in a much tighter fit (mean squared deviation. m.s.d., = 0.197) between experimental and calculated values of log P compared with the other 14 methods for which the m.s.d. ranged between 0.247 and 1.068. The major differentiation between VLOGP and other predictors is that its application domain, called Optimum Prediction Space (OPS), is quantitatively defined, i.e., the structures to which VLOGP model should not be applied for predicting log P can be identified. This process is automated by implementation of VLOGP in the TOPKAT package.

Key Words:

• Log P
• QSAR
• drug partition
• model application domain