Aqueous solubility, predictive model comparisons

The method was tested on a validation set consisting of 205 compounds (41 nonpolar and 164 polar chemicals). In addition, the method was compared to regression methods using S and Kow as the input variables. Results show that this model outperforms and covers a wider range of chemical structures than models based on Kow aqueous solubility, or single MCIs. Meylan et al. (1992) concluded that the MCI-fragment contribution method was clearly superior in the prediction of K(k , although in many cases experimental values of S and Kow were not available and estimated values were used. They presented no direct comparison with other MCI-based estimation methods. 

The benchmarking of methods for physicochemical property predictions is a topic of considerable interest, particularly for the comparison of commercial programs. There were a number of studies and reviews that addressed this problem for prediction of lipophilicity [82-84] aqueous solubility [19,44] as well as other physicochemical property prediction methods [115]. Indeed, since the final purpose of any model development is to predict new data, the benchmarking of different models allows one to compare the advantages of different methodologies. 

Next to ADME phenomena, recent data mining studies also focused on the development or improvement of models predicting physicochemical properties relevant to the field of ADME. Examples are Henry s law constant [92], polar surface area [93], and log P [94]. These models try to overcome limitations of already existing models, see for example SlogP [94] vs. Clogp [95], or aqueous solubility [96], The latter study used more than 2000 compounds selected from the AQUASOL [97] and PHYSOPROP [98] databases. Comparison with a multilinear regression showed clear preference for the neural network. 

We will now consider the case of two predictive models. Model A and Model B, for aqueous solubility. We will use R to compare the performance of these models when tested on 25, 50, and 100 compounds. Listing 9 provides an example of how this comparison can be performed in R. In this listing, we first calculate the Pearson r and the upper and lower 95% confidence intervals for the Pearson r. Table 1.6 and Figure 1.5 show the correlations and associated bar plots. The bar plots show the value of Pearson r for each subset and the associated whiskers show the upper and lower limits of the 95% confidence interval. 

A key question is as follows Can SE and DSE, as an information theoretic approach to descriptor comparison and selection, be applied to accurately classify compoimds or to model physiochemical properties To answer this question, two conceptually different applications of SE and DSE analysis will be discussed here and related to other studies. The first application explores systematic differences between compound sets from synthetic and natural sources." The second addresses the problem of rational descriptor selection to predict the aqueous solubility of synthetic compounds." For these purposes, SE or DSE analysis were carried out, and in both cases, selected descriptors were used to build binary QSAR-like classification models. 

Figure A-23 Solubility of Zr(0H)4(am). Comparison of the data of [72DER] and [66BIL/BRA2] with model prediction for 1 M NH4NO3 using the hydrolysis constants for aqueous species selected in this review, logm (Zr(OH)4(am, fresh), 298,15 K) = - (3.24 0.10), and s(Zr3(OH), NO() = (0.5 0.3) (estimated from the average interaction parameter for trivalent cations (Table B-4, Appendix B). In order to obtain an optimal fit of the data, e( Zr. (OH), NO)) = 1.0 is required.

---