Octanol-water partition: searching for predictive models

The log n-octanol/water partition coefficient (log Po/w) still represents one of the most informative physicochemical parameters available to medicinal chemists. In the present work, principles, methodologies, and parameters are briefly reviewed for a variety of models developed to predict this parameter based on molecular structure. To include the developments of recent years, a total of more than 40 different approaches are mentioned with relevant bibliography within four major categories: group contribution methods, atomic contribution methods, molecular methods, and other physicochemical methods. To underscore once more the utility of this partition coefficient, a comprehensive and reevaluated correlation between log Po/w and in vivo permeability data of rat brain capillaries is included. Most deviants that fell below the trendline are those that have been recently found to be substrates for P-glycoprotein, a multidrug transporter that actively removes them from the brain. Accurate predictions of log Po/w may necessitate many parameters, but there is mounting evidence that molecular size and hydrogen bonding ability can account for a major part of the variance. Our recently developed, molecular size-based approach is reviewed, and it is argued that introduction of three-dimensionality allows the elimination of many empirically derived fragment constants without a significant deterioration of the predictive accuracy. A comparison of predictive power for six different methods on 145 molecules of interest for medicinal chemists is also included.