Log D prediction

Log D predictions are more difficult as most approaches rely on the combination of estimated log P and estimated pK. Obviously, this can lead to error accumulation and errors of 2 log units or more can be found. Some algorithms, however, are designed to learn from experimental data so that the predictions improve over time. 

Considering the failure of available log D methods to predict in-house data and taking into account that such data are usually generated just for a few fixed pH values, a number of companies started to elaborate in-house methods for log D prediction at fixed pH. Up to date several companies have reported development of such methods. For example, Cerep has developed methods to predict log D at pH 7.4 and 6.5 included in their Bio Print package [107], but details of their method are not pubhshed. HQSAR Tripos descriptors were used by Bayer to develop log D models at pH 2.3 and 7.5 using 70000 (qi =0.76, STD =0.60) and 7000 (qi =0.83, STD =0.67) compounds, respectively [108] however, again, no details of the approach were provided. 

The models developed for log D prediction usually aim at being global ones. This, however, does not work on practice. Sheridan et al. [109] noticed that the accuracy of log D prediction of molecules decreased approximately 2-3 times (RMSE = 0.75 versus 1.5-2) as the similarity of the test molecule to the molecules in the training set (using Dice definition with the atom pair descriptors) changed from 1 to 0 (most to least similar). Thus, if a test set molecule had a very similar molecule in the training set, it was possible to accurately predict its log D value. A detailed overview of state of the art methods to access the same problem was published elsewhere [117]. 

A low accuracy of models for prediction of log D at any pH would not encourage the use of these models for practical applications in industry. Thus, it is likely that the methods for log D prediction at fixed pH that are developed in house by pharmaceutical companies will dominate in industry. However, log D measurements... 

From these equations it is possible to predict the effective lipophilicity (log D) of an acidic or basic compound at any pH value. The data required in order to use the relationship in this way are the intrinsic lipophilicity (log P), the dissociation constant (pKa) and the pH of the aqueous phase. The overaU effect of these relahonships is the effechve hpophilicity of a compound, at physiological pH, is approximately the log P value minus one unit of hpophilicity, for every unit of pH the pKa value is below (for acids) and above (for bases) pH 7.4. Obviously for compounds with mul-hfunchonal ionizable groups the relahonship between log P and log D, as weU as log D as a function of pH become more complex [65, 68, 70]. For diprotic molecules there are already 12 different possible shapes of log D-pH plots. 

Bruneau, P., McElroy, N. R. Log D " modeling using Bayesian regularised neural networks. Assessment and correction of errors of prediction. 

Log P and log D can be experimentally measured and computationally calculated. Both measurements and calculations can be made by a variety of methods, most of which are quite simple to perform (see following chapters). Our experience recommends, if possible, the use of both procedures. In fact the combination of theory (i.e. how things should be) with practice (i.e. how things are) enables both a better set-up of experiments and the identification of the best predictive method to be used for the chosen dataset of compounds. 

Lipophilicity is intuitively felt to be a key parameter in predicting and interpreting permeability and thus the number of types of lipophilicity systems under study has grown enormously over the years to increase the chances of finding good mimics of biomembrane models. However, the relationship between lipophilicity descriptors and the membrane permeation process is not clear. Membrane permeation is due to two main components the partition rate constant between the lipid leaflet and the aqueous environment and the flip-flop rate constant between the two lipid leaflets in the bilayer [13]. Since the flip-flop is supposed to be rate limiting in the permeation process, permeation is determined by the partition coefficient between the lipid and the aqueous phase (which can easily be determined by log D) and the flip-flop rate constant, which may or may not depend on lipophilicity and if it does so depend, on which lipophilicity scale should it be based ... 

To sum up, lipophilicity is only one component of permeability, and thus any relationships found between passive permeation and log D] are reliable for the investigated series of compounds, but cannot be used to make general predictions. 

Eros, D., Kovesdi, L, Orfi, L., Takacs-Novak, K., Acsady, G., Keril, G. Reliability of log P predictions based on calculated molecular descriptors a critical review. Curr. Med. Chem. 2002, 9,1819-1829. 

Tetko, I. V., Bruneau, P. Application of ALOGPS to predict 1-octanol/water distribution coefficients, log P, and log D, of AstraZeneca in-house database. 

While there are plenty of methods to predict 1-octanol-water partition coefficients, logP (see Chapters 14 and 15), the number of approaches to predict 1-octanol-water distribution coefficients is rather limited. This is due to a lower availability of log D data and, in general, higher computational complexity of this property compared to that of log P. The approaches to predict log D can be roughly classified into two major categories (i) calculation of log D at an arbitrary pH and (ii) calculation of log D at a fixed pH. 

Actually, why do we need log D models Why can t we use just log P models One of the main requirements for prediction of octanol-water coefficients is to optimize bioavailability of chemical compounds. During the absorption process the... 

In order to be appUed the models should be predictive. Unfortunately, the models frequently fail and demonstrate significantly lower prediction ability compared to the estimated one, when they are applied to new unseen data [100-103, 106]. One of the main reasons for such failures can be the lack of available experimental data and difficulties in calculating log D, as discussed in Section 16.4.2. Another problem of low prediction ability of log D models can be attributed to different chemical diversity of molecules in the in-house databases compared to the training sets used to develop the programs. 

However, there is still a strong need to develop new methods that will be able to quantitatively or at least qualitatively estimate the prediction accuracy of log D models. Such models will allow the computational chemist to distinguish reliable versus nonreliable predictions and to decide whether the available model is sufficiently accurate or whether experimental measurements should be provided. For example, when applying ALOGPS in the LIB RARY model it was possible to predict more than 50% and 30% compounds with an accuracy of MAE <0.35 for Pfizer and AstraZeneca collections, respectively [117]. This precision approximately corresponds to the experimental accuracy, s=0.4, of potentiometric lipophilicity determinations [15], Thus, depending on the required precision, one could skip experimental measurements for some of the accurately predicted compounds. 

The un-ionized form is assumed to be sufficiently lipophilic to traverse membranes in the pH-partition hypothesis. If it were not, no transfer could be predicted, irrespective of pH. The lipophilicity of compounds is experimentally determined as the partition coefficient (log P) or distribution coefficient (log D) [16]. The partition coefficient is the ratio of concentrations of the neutral species between aqueous and nonpolar phases, while the distribution coefficient is the ratio of all species between aqueous and nonpolar phases [17,18],... 

The mechanistic simulation ACAT model was modified to account automatically for the change in small intestinal and colon k as a function of the local (pH-dependent) log D of the drug molecule. The rank order of %HIA from GastroPlus was directly compared with rank order experimental %HIA with this correction for the log D of each molecule in each of the pH environments of the small intestine. A significant Spearman rank correlation coefficient for the mechanistic simulation-based method of 0.58 (p < 0.001) was found. The mechanistic simulation produced 71% of %HIA predictions within 25% of the experimental values. 

Combination of several descriptors believed to be important for oral absorption have been used in various multivariate analysis studies [26]. The general trend is that a combination of size/shape and a hydrogen bond descriptor, sometimes in combination with log D, has good predictive value. At present such models do not account for the biological function of the membrane, such as P-gp-mediated efflux. 

The ADME portion of the BioPrint profile is made up of a panel of in vitro assays chosen for their potential to predict in vivo pharmacokinetics (Table 2). Some of the in vitro assays measure properties that contribute to the in vivo bioavailability of the new drug candidate. These include aqueous solubility, log D (octanol), and log D (cyclohexane), physico-chemical... 

A growing list of predictive QSAR models derived from BioPrint data e.g. solubility, log D (pH = 7.4, pH = 6.5), Cyp2D6 inhibition, permeability (A B, B A, passive), and more than 30 target-related models. 

---