VolSurf descriptor

A, B and V are constant for a given solute (Eig. 12.4 shows the value of A, 0.78, for atenolol). This means that the balance between intermolecular forces varies with the system investigated as would be expected from a careful reading of Section 12.1.1.3. This can also be demonstrated by using a completely different approach to factorize log P, i.e. a computational method based on molecular interaction fields [10]. Volsurf descriptors [11] have been used to calculate log P of neutral species both in n-octanol-water and in alkane-water [10]. 

Thus, a molecule can be characterized in terms of its potential hydrogen bonding, polar, hydrophobic and ionic interactions in 3D space. The size and the spatial distribution of these molecular interaction contours is translated into a quantitative scheme, the VolSurf descriptors, without the need to align the molecules in 3D space [8, 9] (Fig. 17.1). 

Fig. 17.1. Multivariate characterization with VolSurf descriptors. Molecular Interaction Fields (MIF shaded areas) are computed from the 3D-molecular structure. MIFs are transformed in a table of descriptors, and statistical multivariate analysis is performed.

The molecular descriptors refer to the molecular size and shape, to the size and shape of hydrophilic and hydrophobic regions, and to the balance between them. Hydrogen bonding, amphiphilic moments, critical packing parameters are other useful descriptors. The VolSurf descriptors have been presented and explained in detail elsewhere [8]. The VolSurf descriptors encode physico-chemical properties and, therefore, allow both for a design in the physico-chemical property space in order to rationally modulate pharmacokinetic properties, and for establishing quantitative structure-property relationships (QSPR). 

Calculated molecular properties from 3D molecular fields of interaction energies are a novel approach to correlate 3D molecular structures with pharmacodynamic, pharmacokinetic and physico-chemical properties. The novel VolSurf descriptors quantitatively characterize size, shape, polarity, hydrophobicity and the balance between them. 

Cabrera et al. [50] modeled a set of 163 drugs using TOPS-MODE descriptors with a linear discriminant model to predict p-glycoprotein efflux. Model accuracy was 81% for the training set and 77.5% for a validation set of 40 molecules. A "combinatorial QSAR" approach was used by de Lima et al. [51] to test multiple model types (kNN, decision tree, binary QSAR, SVM) with multiple descriptor sets from various software packages (MolconnZ, Atom Pair, VoSurf, MOE) for the prediction of p-glycoprotein substrates for a dataset of 192 molecules. Best overall performance on a test set of 51 molecules was achieved with an SVM and AP or VolSurf descriptors (81% accuracy each). 

Percent renal clearance was modeled for a set of 130 compounds from the literature using partial least squares applied to 3-D VolSurf or 2-D Molconn-Z descriptors [74]. The model based on VolSurf descriptors gave the best prediction... 

Apart from pharmacophore-based approaches, a variety of methods were applied to decipher important ligand features of PXR activation. VolSurf descriptor-based partial least squares (PLS) regression-based models pointed toward amide responsive regions that implicated good acceptor abilities as key variables [33]. 

These descriptors have been reported in the literature to correlate with bioavailability, blood-brain partitioning, membrane transport and other properties [156-159]. They are also correlated to relevant physicochemical properties and were also successfully applied to many internal and public data. Eor example, we derived PLS models [160] using 72 VolSurf descriptors for human serum albumin (HSA) binding using 95 drugs on a... 

Fig. 14.5 Computation of VolSurf descriptors [155, 156] derived from GRID molecular interaction fields. Interactions of the example molecule with a water and dry probe at different contour levels are used to compute a vector of 72 volume-, size- and surface-based descriptors.

This agrees to internal VolSurf models derived for PAMPA membrane transport [163] to understand passive transcellular transport across membranes. One of our internal models based on 29 compounds characterized by immobilized artificial membrane chromatography by Salminen etal. ]164] shows an of 0.81 and = 0.70 for two PLS components derived using VolSurf descriptors. This is one of the rare examples where ionized starting molecules led to slightly better PLS statistics, while the general chemical interpretation is not affected. 

Mannhold, R., Berellini, G., Carosati, E. and Benedetti, P. (2006) Use ofMlF-based VolSurf descriptors in physicochemical and pharmacokinetic studies, in Molecular Interaction Fields (ed. G. Crudani), Wiley-VCH, Weinheim,... 

Fig. 2. Computation of Volsurf descriptors (Cruciani et al. 2000a) derived from GRID molecular interaction fields. For any molecule, interactions with GRID water and dry probes at different energy levels are used for contouring. Those levels serve to compute vectors of 72 volume, size, and surface related descriptors.

Fig. 3. Correlation of VolSurf descriptors with human intestinal absorption using multivariate statistics (PLS) based on 20 drug molecules as reported by Cuba et al. (2000). The PLS plot (ul versus tl) and the corresponding PLS coefficient plot is shown. Different interaction pattern with the GRID water probe are displayed for the orally available nordiazepam (left) versus the large area for the non-available...

Volsurf descriptors were also successfully used to build a classification model for predicting blood-brain barrier (BBB) permeability for drug-like molecules... 

Another modification of the VolSurf approach to analyze molecular interaction fields was described by Pastor et al. (2000). The same information previously been used to compute VolSurf descriptors now served to compute GRIND descriptors. In contrast to VolSurf, these descriptors capture information about pharmacophoric groups and their distances within individual molecules. These descriptors represent favorable interaction energy regions where groups... 

While the majority of published models are based on a limited number of drug molecules, especially the study of Zhao et al. (2001) provides the most extensive compilation from available literature data and a statistical model derived from those using Abraham descriptors. We used this carefully selected dataset to build a quantitative model for human intestinal absorption employing VolSurf descriptors (see Cruciani et al. 2000). 

This procedure led to a predictive PLS model for 40 VolSurf descriptors and 4 relevant PLS components with an crossvalidated r2(cv) value after leave-one-out crossvalidation of 0.662 and a conventional r2 value... 

Fig. 5. Correlation of VolSurf descriptors with human intestinal absorption for 169 drug molecules. Left. Predicted versus experimental %HIA (human intestinal absorption) from final 4-component PLS model. Right PLS loadings showing the importance of VolSurf descriptors to the prediction of human intestinal absorption.

The correlation of VolSurf descriptors to the human intestinal absorption for 169 drugs is shown in Figure 5 from the final 4-component PLS model. On the left panel the plot of experimental versus predicted % HIA is displayed. Although the data points do not fall onto a straight line, the model clearly is able to discriminate between compounds with high, medium or low intestinal absorption. The analysis of PLS coefficients from this model allows for a chemical interpretation of... 

CRITICAL ASSESSMENT OF THE METHOD VolSurf descriptors are able to predict absorption for a diverse set of drugs. The presented model is derived using a consistent frame of relevant chemically interpretable descriptors, which find applications in different local and general models. However, absorption is not only controlled by passive membrane permeability. There are other factors influencing in vivo human absorption namely the in vivo dissolution rate in small intestinal fluid and the dose used for the human study. Furthermore, active transport or efflux mechanisms are difficult to rule out but can only be partially monitored by in vitro experiments. These important pieces of information should be known before any QSAR analysis is attempted on human absorption. This lack of consistent information throughout the literature is difficult to overcome, in particular for human studies. Hence, this study for the dataset from Zhao et al. (2001) provides a reasonable attempt to address these problems to carefully selecting members of the final dataset. 

The predictivity of this model to external compounds was further evaluated by splitting the dataset into a training set of 83 compounds and a test set of 10 molecules. One approach was to use the test set from Colmenarejo et al. (2001 model B), while another test set was generated using statistical design after a principal component analysis (PCA) on VolSurf descriptors (model A). The design was done by selecting... 

The set of 51 benzamidine-based thrombin inhibitors was taken from the study of Sugano et al. (2000). Experimental rat everted sac permeabilities were expressed as log(ES A) values.2 The experimental permeability in this assay is expressed as ratio of outer (mucosal side) concentration of the drug and inner (serosal side) concentration after 1 h incubation of the everted sac of rat small intestine. All molecules were treated in their neutral form and converted into their 3D structures using CORINA (Sadowski et al. 1992). From GRID molecular interaction fields for water, dry, and carbonyl oxygen probes, a set of 72 VolSurf descriptors (Cruciani et al. 2000) was computed and analysed as described above. 

This procedure led to a predictive 4 component PLS model for 72 VolSurf descriptors and 51 thrombin inhibitors. A crossvalidated r2(cv) value of 0.599 after leave-one-out crossvalidation and a conventional r2 value of 0.812 were obtained. Statistical validation using leave-two-out and leave-multiple-groups-out crossvalidation procedures underscores the significance of the final model. The graph of experimental versus calculated log(ESA) permeability values is shown in Figure 8 on the left. The overall model quality corresponds to the model reported by Sugano etal. (2000). 

---