A new approach for predicting the lipophilicity (log P), solubility (log Sw), and oral absorption of drugs in humans (FA) is described. It is based on structural and physicochemical similarity and is realized in the software program SLIPPER-2001. Calculated and experimental values of log P, log Sw, and FA for 42 drugs were used to demonstrate the predictive power of the program. Reliable results were obtained for simple compounds, for complex chemicals, and for drugs. Thus, the principle of "similar compounds display similar properties" together with estimating incremental changes in properties by using differences in physicochemical parameters results in "structure - property " predictive models even in the absence of a precise understanding of the mechanisms involved.

A detailed QSPR investigation of the water solubility (logS) of 1063 solid neutral chemicals, agrochemicals, drugs and prodrugs has been carried out. The application of the ™General Solubility Equation∫ of Yalkowsky et al. resulted in a correlation between experimental and calculated solubility with rather modest statistic criteria. It was found that 191 compounds (18%) have calculated logS with deviations above one logarithmic unit. A comparison of experimental values with those calculated by an equation previously derived for liquid chemicals demonstrates that a part of the total set of compounds containing certain substructures such as chloroalkyls, phenyls, biphenyls, nitrogen (acyclic and in cycles), benzanthracenes, phenols, and chemicals with ether, ester, and N,N-disubstituted carboxamide groups obey to the equation for liquids. Not unexpectedly, the other part of compounds containing both strong H-bond acceptor and donor groups is essentially less soluble than calculated for compounds in the liquid state. It is proposed that this low solubility is connected with a specific H-bond association of these compounds in their crystal lattice.Two approaches were tested for the derivation of quantitative models for prediction of the solubility of solid chemicals and drugs. The first approach is based on the QSPR for liquids extended by several indicator variables for different functional groups. The second approach is based on the combination of chemical similarity and traditional QSAR techniques. In the framework of this approach different numbers of structural and physicochemical neighbors of a compound-of-interest were considered together with the corresponding HYBOT descriptor data. This method enables to calculate the solubility of a compound-of-interest by using the solubility of nearest neighbor compounds and the difference between descriptor data of those neighbors and the corresponding data of the compound-of-interest. It was found that the use of nearest neighbor compounds with high Tanimoto index value (i.e. good structural similarity) and close H-bond acceptor and donor factors (good physicochemical similarity) ensures good prediction of water solubility with average absolute errors on the level of the error of experimental logS determination.

Detailed critical analysis of publications devoted to QSPR of aqueous solubility is presented in the review with discussion of four types of aqueous solubility (three different thermodynamic solubilities with unknown solute structure, intrinsic solubility, solubility in physiological media at pH=7.4 and kinetic solubility), variety of molecular descriptors (from topological to quantum chemical), traditional statistical and machine learning methods as well as original QSPR models.

32 Quantitative Structure-Property Relationship (QSPR) models were constructed for prediction of aqueous intrinsic solubility of liquid and crystalline chemicals. Data sets contained 1022 liquid and 2615 crystalline compounds. Multiple Linear Regression (MLR), Support Vector Machine (SVM) and Random Forest (RF) methods were used to construct global models, and k-nearest neighbour (kNN), Arithmetic Mean Property (AMP) and Local Regression Property (LoReP) were used to construct local models. A set of the best QSPR models was obtained: for liquid chemicals with RMSE (root mean square error) of prediction in the range 0.50-0.60 log unit; for crystalline chemicals 0.80-0.90 log unit. In the case of global models the large number of descriptors makes mechanistic interpretation difficult. The local models use only one or two descriptors, so that a medicinal chemist working with sets of structurally-related chemicals can readily estimate their solubility. However, construction of stable local models requires the presence of closely related neighbours for each chemical considered. It is probable that a consensus of global and local QSPR models will be the optimal approach for construction of stable predictive QSPR models with mechanistic interpretation.

QSPR analyses of a data set containing experimental partition coefficients in the three systems octanol-water, water-gas, and octanol-gas for 98 chemicals have shown that it is possible to calculate any partition coefficient in the system 'gas phase/octanol/water' by three different approaches: (1) from experimental partition coefficients obtained in the corresponding two other subsystems. However, in many cases these data may not be available. Therefore, a solution may be approached (2), a traditional QSPR analysis based on e.g. HYBOT descriptors (hydrogen bond acceptor and donor factors, SigmaCa and SigmaCd, together with polarisability alpha, a steric bulk effect descriptor) and supplemented with substructural indicator variables. (3) A very promising approach which is a combination of the similarity concept and QSPR based on HYBOT descriptors. In this approach observed partition coefficients of structurally nearest neighbours of a compound-of-interest are used. In addition, contributions arising from differences in alpha, SigmaCa, and SigmaCd values between the compound-of-interest and its nearest neighbour(s), respectively, are considered. In this investigation highly significant relationships were obtained by approaches (1) and (3) for the octanol/gas phase partition coefficient (log Log).